diff --git a/dev/yes-no-maybe-vision/train.ipynb b/dev/yes-no-maybe-vision/train.ipynb
index 46d29c72..0a5e782d 100644
--- a/dev/yes-no-maybe-vision/train.ipynb
+++ b/dev/yes-no-maybe-vision/train.ipynb
@@ -2,7 +2,7 @@
  "cells": [
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 1,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -12,9 +12,30 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 2,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "text/html": [
+       "<style>\n",
+       ".cell-output-ipywidget-background {\n",
+       "    background-color: transparent !important;\n",
+       "}\n",
+       ":root {\n",
+       "    --jp-widgets-color: var(--vscode-editor-foreground);\n",
+       "    --jp-widgets-font-size: var(--vscode-editor-font-size);\n",
+       "}  \n",
+       "</style>\n"
+      ],
+      "text/plain": [
+       "<IPython.core.display.HTML object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
    "source": [
     "%%html\n",
     "<style>\n",
@@ -30,9 +51,207 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 3,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Skipping import of cpp extensions due to incompatible torch version 2.7.1+cu126 for torchao version 0.14.0         Please see GitHub issue #2919 for more info\n",
+      "\u001b[34m\u001b[1mwandb\u001b[0m: Currently logged in as: \u001b[33mbhilton\u001b[0m (\u001b[33mwandb\u001b[0m) to \u001b[32mhttps://api.wandb.ai\u001b[0m. Use \u001b[1m`wandb login --relogin`\u001b[0m to force relogin\n"
+     ]
+    },
+    {
+     "data": {
+      "text/html": [
+       "Tracking run with wandb version 0.21.0"
+      ],
+      "text/plain": [
+       "<IPython.core.display.HTML object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "Run data is saved locally in <code>/home/sky/sky_workdir/dev/yes-no-maybe-vision/wandb/run-20251022_222913-009</code>"
+      ],
+      "text/plain": [
+       "<IPython.core.display.HTML object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "Resuming run <strong><a href='https://wandb.ai/wandb/yes-no-maybe-vision/runs/009' target=\"_blank\">009</a></strong> to <a href='https://wandb.ai/wandb/yes-no-maybe-vision' target=\"_blank\">Weights & Biases</a> (<a href='https://wandb.me/developer-guide' target=\"_blank\">docs</a>)<br>"
+      ],
+      "text/plain": [
+       "<IPython.core.display.HTML object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       " View project at <a href='https://wandb.ai/wandb/yes-no-maybe-vision' target=\"_blank\">https://wandb.ai/wandb/yes-no-maybe-vision</a>"
+      ],
+      "text/plain": [
+       "<IPython.core.display.HTML object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       " View run at <a href='https://wandb.ai/wandb/yes-no-maybe-vision/runs/009' target=\"_blank\">https://wandb.ai/wandb/yes-no-maybe-vision/runs/009</a>"
+      ],
+      "text/plain": [
+       "<IPython.core.display.HTML object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "INFO 10-22 22:29:16 [__init__.py:235] Automatically detected platform cuda.\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Skipping import of cpp extensions due to incompatible torch version 2.7.1+cu126 for torchao version 0.14.0         Please see GitHub issue #2919 for more info\n",
+      "/home/sky/sky_workdir/src/art/__init__.py:10: UserWarning: WARNING: Unsloth should be imported before transformers, peft to ensure all optimizations are applied. Your code may run slower or encounter memory issues without these optimizations.\n",
+      "\n",
+      "Please restructure your imports with 'import unsloth' at the top of your file.\n",
+      "  import unsloth  # type: ignore # noqa: F401\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "🦥 Unsloth: Will patch your computer to enable 2x faster free finetuning.\n",
+      "INFO 10-22 22:29:22 [__init__.py:235] Automatically detected platform cuda.\n",
+      "🦥 Unsloth Zoo will now patch everything to make training faster!\n",
+      "==((====))==  Unsloth 2025.10.8: Fast Qwen3_Vl patching. Transformers: 4.57.1. vLLM: 0.10.0.\n",
+      "   \\\\   /|    NVIDIA H200. Num GPUs = 1. Max memory: 139.811 GB. Platform: Linux.\n",
+      "O^O/ \\_/ \\    Torch: 2.7.1+cu126. CUDA: 9.0. CUDA Toolkit: 12.6. Triton: 3.3.1\n",
+      "\\        /    Bfloat16 = TRUE. FA [Xformers = 0.0.31. FA2 = False]\n",
+      " \"-____-\"     Free license: http://github.com/unslothai/unsloth\n",
+      "Unsloth: Fast downloading is enabled - ignore downloading bars which are red colored!\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "INFO:unsloth_zoo.log: Unsloth: Patching vLLM\n",
+      "WARNING:unsloth_zoo.log: Unsloth: Switching from Unsloth dynamic quant to normal quant since\n",
+      "we do not yet support fast inference for unsloth/qwen3-vl-8b-instruct-unsloth-bnb-4bit\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "INFO 10-22 22:29:34 [vllm_utils.py:694] Unsloth: Patching vLLM v1 graph capture\n",
+      "INFO 10-22 22:29:34 [vllm_utils.py:722] Unsloth: Patching vLLM v0 graph capture\n",
+      "Unsloth: Vision model detected, setting approx_max_num_seqs to 1\n",
+      "Unsloth: vLLM loading unsloth/qwen3-vl-8b-instruct-bnb-4bit with actual GPU utilization = 78.66%\n",
+      "Unsloth: Your GPU has CUDA compute capability 9.0 with VRAM = 139.81 GB.\n",
+      "Unsloth: Using conservativeness = 1.0. Chunked prefill tokens = 2048. Num Sequences = 1.\n",
+      "Unsloth: vLLM's KV Cache can use up to 103.57 GB. Also swap space = 6 GB.\n",
+      "Unsloth: Not an error, but `device` is not supported in vLLM. Skipping.\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "`torch_dtype` is deprecated! Use `dtype` instead!\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "INFO 10-22 22:29:41 [config.py:1604] Using max model len 32768\n",
+      "WARNING 10-22 22:29:41 [arg_utils.py:1511] --enable-prefix-caching is not supported for multimodal models in V0 and has been disabled.\n",
+      "INFO 10-22 22:29:41 [config.py:2434] Chunked prefill is enabled with max_num_batched_tokens=32768.\n",
+      "Unsloth: vLLM Bitsandbytes config using kwargs = {'load_in_8bit': False, 'load_in_4bit': True, 'bnb_4bit_compute_dtype': 'bfloat16', 'bnb_4bit_quant_storage': 'uint8', 'bnb_4bit_quant_type': 'nf4', 'bnb_4bit_use_double_quant': True, 'llm_int8_enable_fp32_cpu_offload': False, 'llm_int8_has_fp16_weight': False, 'llm_int8_skip_modules': ['embed_tokens', 'embedding', 'lm_head', 'multi_modal_projector', 'merger', 'modality_projection', 'router', 'visual', 'vision_tower'], 'llm_int8_threshold': 6.0}\n",
+      "INFO 10-22 22:29:41 [llm_engine.py:228] Initializing a V0 LLM engine (v0.10.0) with config: model='unsloth/qwen3-vl-8b-instruct-bnb-4bit', speculative_config=None, tokenizer='unsloth/qwen3-vl-8b-instruct-bnb-4bit', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, override_neuron_config={}, tokenizer_revision=None, trust_remote_code=False, dtype=torch.bfloat16, max_seq_len=32768, download_dir=None, load_format=bitsandbytes, tensor_parallel_size=1, pipeline_parallel_size=1, disable_custom_all_reduce=False, quantization=bitsandbytes, enforce_eager=False, kv_cache_dtype=auto,  device_config=cuda, decoding_config=DecodingConfig(backend='auto', disable_fallback=False, disable_any_whitespace=False, disable_additional_properties=False, reasoning_backend=''), observability_config=ObservabilityConfig(show_hidden_metrics_for_version=None, otlp_traces_endpoint=None, collect_detailed_traces=None), seed=0, served_model_name=unsloth/qwen3-vl-8b-instruct-bnb-4bit, num_scheduler_steps=1, multi_step_stream_outputs=True, enable_prefix_caching=False, chunked_prefill_enabled=True, use_async_output_proc=True, pooler_config=None, compilation_config={\"level\":0,\"debug_dump_path\":\"\",\"cache_dir\":\"\",\"backend\":\"inductor\",\"custom_ops\":[],\"splitting_ops\":[],\"use_inductor\":true,\"compile_sizes\":[],\"inductor_compile_config\":{\"epilogue_fusion\":true,\"max_autotune\":false,\"shape_padding\":true,\"trace.enabled\":false,\"triton.cudagraphs\":true,\"debug\":false,\"dce\":true,\"memory_planning\":true,\"coordinate_descent_tuning\":false,\"trace.graph_diagram\":false,\"compile_threads\":32,\"group_fusion\":true,\"disable_progress\":false,\"verbose_progress\":true,\"triton.multi_kernel\":0,\"triton.use_block_ptr\":true,\"triton.enable_persistent_tma_matmul\":true,\"triton.autotune_at_compile_time\":false,\"triton.cooperative_reductions\":false,\"cuda.compile_opt_level\":\"-O2\",\"cuda.enable_cuda_lto\":true,\"combo_kernels\":false,\"benchmark_combo_kernel\":true,\"combo_kernel_foreach_dynamic_shapes\":true,\"enable_auto_functionalized_v2\":false},\"inductor_passes\":{},\"use_cudagraph\":true,\"cudagraph_num_of_warmups\":1,\"cudagraph_capture_sizes\":[1],\"cudagraph_copy_inputs\":false,\"full_cuda_graph\":true,\"max_capture_size\":1,\"local_cache_dir\":null}, use_cached_outputs=False, \n",
+      "INFO 10-22 22:29:46 [cuda.py:398] Using Flash Attention backend.\n",
+      "INFO 10-22 22:29:46 [parallel_state.py:1102] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, TP rank 0, EP rank 0\n",
+      "INFO 10-22 22:29:46 [model_runner.py:1083] Starting to load model unsloth/qwen3-vl-8b-instruct-bnb-4bit...\n",
+      "WARNING 10-22 22:29:46 [utils.py:185] Qwen3VLForConditionalGeneration has no vLLM implementation, falling back to Transformers implementation. Some features may not be supported and performance may not be optimal.\n",
+      "INFO 10-22 22:29:46 [transformers.py:421] Using Transformers backend.\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "`torch_dtype` is deprecated! Use `dtype` instead!\n"
+     ]
+    },
+    {
+     "ename": "RuntimeError",
+     "evalue": "vLLM currently does not support BNB quantization for",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
+      "\u001b[0;31mAssertionError\u001b[0m                            Traceback (most recent call last)",
+      "File \u001b[0;32m~/sky_workdir/.venv/lib/python3.10/site-packages/unsloth_zoo/vllm_utils.py:1763\u001b[0m, in \u001b[0;36mload_vllm\u001b[0;34m()\u001b[0m\n\u001b[1;32m   1762\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m use_async:\n\u001b[0;32m-> 1763\u001b[0m     llm \u001b[38;5;241m=\u001b[39m AsyncLLMEngine\u001b[38;5;241m.\u001b[39mfrom_engine_args(AsyncEngineArgs(\u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mengine_args))\n\u001b[1;32m   1764\u001b[0m \u001b[38;5;28;01melif\u001b[39;00m use_engine:\n",
+      "File \u001b[0;32m~/sky_workdir/src/art/unsloth/state.py:74\u001b[0m, in \u001b[0;36m_from_engine_args\u001b[0;34m()\u001b[0m\n\u001b[1;32m     71\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[38;5;21m_from_engine_args\u001b[39m(\n\u001b[1;32m     72\u001b[0m     engine_args: AsyncEngineArgs, \u001b[38;5;241m*\u001b[39margs: Any, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs: Any\n\u001b[1;32m     73\u001b[0m ) \u001b[38;5;241m-\u001b[39m\u001b[38;5;241m>\u001b[39m AsyncLLMEngine:\n\u001b[0;32m---> 74\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m from_engine_args(\n\u001b[1;32m     75\u001b[0m         replace(engine_args, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mconfig\u001b[38;5;241m.\u001b[39mget(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mengine_args\u001b[39m\u001b[38;5;124m\"\u001b[39m, {})), \u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs\n\u001b[1;32m     76\u001b[0m     )\n",
+      "File \u001b[0;32m~/sky_workdir/.venv/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py:653\u001b[0m, in \u001b[0;36mfrom_engine_args\u001b[0;34m()\u001b[0m\n\u001b[1;32m    651\u001b[0m     async_engine_cls \u001b[38;5;241m=\u001b[39m V1AsyncLLMEngine\n\u001b[0;32m--> 653\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m async_engine_cls\u001b[38;5;241m.\u001b[39mfrom_vllm_config(\n\u001b[1;32m    654\u001b[0m     vllm_config\u001b[38;5;241m=\u001b[39mvllm_config,\n\u001b[1;32m    655\u001b[0m     start_engine_loop\u001b[38;5;241m=\u001b[39mstart_engine_loop,\n\u001b[1;32m    656\u001b[0m     usage_context\u001b[38;5;241m=\u001b[39musage_context,\n\u001b[1;32m    657\u001b[0m     stat_loggers\u001b[38;5;241m=\u001b[39mstat_loggers,\n\u001b[1;32m    658\u001b[0m     disable_log_stats\u001b[38;5;241m=\u001b[39mengine_args\u001b[38;5;241m.\u001b[39mdisable_log_stats,\n\u001b[1;32m    659\u001b[0m     disable_log_requests\u001b[38;5;241m=\u001b[39mengine_args\u001b[38;5;241m.\u001b[39mdisable_log_requests,\n\u001b[1;32m    660\u001b[0m )\n",
+      "File \u001b[0;32m~/sky_workdir/.venv/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py:626\u001b[0m, in \u001b[0;36mfrom_vllm_config\u001b[0;34m()\u001b[0m\n\u001b[1;32m    624\u001b[0m \u001b[38;5;250m\u001b[39m\u001b[38;5;124;03m\"\"\"Create an AsyncLLMEngine from the EngineArgs.\"\"\"\u001b[39;00m\n\u001b[0;32m--> 626\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mcls\u001b[39m(\n\u001b[1;32m    627\u001b[0m     vllm_config\u001b[38;5;241m=\u001b[39mvllm_config,\n\u001b[1;32m    628\u001b[0m     executor_class\u001b[38;5;241m=\u001b[39m\u001b[38;5;28mcls\u001b[39m\u001b[38;5;241m.\u001b[39m_get_executor_cls(vllm_config),\n\u001b[1;32m    629\u001b[0m     start_engine_loop\u001b[38;5;241m=\u001b[39mstart_engine_loop,\n\u001b[1;32m    630\u001b[0m     log_requests\u001b[38;5;241m=\u001b[39m\u001b[38;5;129;01mnot\u001b[39;00m disable_log_requests,\n\u001b[1;32m    631\u001b[0m     log_stats\u001b[38;5;241m=\u001b[39m\u001b[38;5;129;01mnot\u001b[39;00m disable_log_stats,\n\u001b[1;32m    632\u001b[0m     usage_context\u001b[38;5;241m=\u001b[39musage_context,\n\u001b[1;32m    633\u001b[0m     stat_loggers\u001b[38;5;241m=\u001b[39mstat_loggers,\n\u001b[1;32m    634\u001b[0m )\n",
+      "File \u001b[0;32m~/sky_workdir/.venv/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py:581\u001b[0m, in \u001b[0;36m__init__\u001b[0;34m()\u001b[0m\n\u001b[1;32m    580\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mlog_requests \u001b[38;5;241m=\u001b[39m log_requests\n\u001b[0;32m--> 581\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mengine \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_engine_class(\u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs)\n\u001b[1;32m    583\u001b[0m \u001b[38;5;66;03m# This ensures quick processing of request outputs\u001b[39;00m\n\u001b[1;32m    584\u001b[0m \u001b[38;5;66;03m# so the append to asyncio queues is not delayed,\u001b[39;00m\n\u001b[1;32m    585\u001b[0m \u001b[38;5;66;03m# especially for multi-step.\u001b[39;00m\n",
+      "File \u001b[0;32m~/sky_workdir/.venv/lib/python3.10/site-packages/vllm/engine/async_llm_engine.py:265\u001b[0m, in \u001b[0;36m__init__\u001b[0;34m()\u001b[0m\n\u001b[1;32m    264\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[38;5;21m__init__\u001b[39m(\u001b[38;5;28mself\u001b[39m, \u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs):\n\u001b[0;32m--> 265\u001b[0m     \u001b[38;5;28msuper\u001b[39m()\u001b[38;5;241m.\u001b[39m\u001b[38;5;21m__init__\u001b[39m(\u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs)\n",
+      "File \u001b[0;32m~/sky_workdir/.venv/lib/python3.10/site-packages/vllm/engine/llm_engine.py:263\u001b[0m, in \u001b[0;36m__init__\u001b[0;34m()\u001b[0m\n\u001b[1;32m    259\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39minput_preprocessor \u001b[38;5;241m=\u001b[39m InputPreprocessor(\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mmodel_config,\n\u001b[1;32m    260\u001b[0m                                             \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mtokenizer,\n\u001b[1;32m    261\u001b[0m                                             mm_registry)\n\u001b[0;32m--> 263\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mmodel_executor \u001b[38;5;241m=\u001b[39m executor_class(vllm_config\u001b[38;5;241m=\u001b[39mvllm_config)\n\u001b[1;32m    265\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mmodel_config\u001b[38;5;241m.\u001b[39mrunner_type \u001b[38;5;241m!=\u001b[39m \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mpooling\u001b[39m\u001b[38;5;124m\"\u001b[39m:\n",
+      "File \u001b[0;32m~/sky_workdir/.venv/lib/python3.10/site-packages/vllm/executor/executor_base.py:53\u001b[0m, in \u001b[0;36m__init__\u001b[0;34m()\u001b[0m\n\u001b[1;32m     52\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mobservability_config \u001b[38;5;241m=\u001b[39m vllm_config\u001b[38;5;241m.\u001b[39mobservability_config\n\u001b[0;32m---> 53\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_init_executor()\n\u001b[1;32m     54\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mis_sleeping \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mFalse\u001b[39;00m\n",
+      "File \u001b[0;32m~/sky_workdir/.venv/lib/python3.10/site-packages/vllm/executor/uniproc_executor.py:49\u001b[0m, in \u001b[0;36m_init_executor\u001b[0;34m()\u001b[0m\n\u001b[1;32m     48\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcollective_rpc(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124minit_device\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n\u001b[0;32m---> 49\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcollective_rpc(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mload_model\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n",
+      "File \u001b[0;32m~/sky_workdir/.venv/lib/python3.10/site-packages/vllm/executor/uniproc_executor.py:58\u001b[0m, in \u001b[0;36mcollective_rpc\u001b[0;34m()\u001b[0m\n\u001b[1;32m     57\u001b[0m     kwargs \u001b[38;5;241m=\u001b[39m {}\n\u001b[0;32m---> 58\u001b[0m answer \u001b[38;5;241m=\u001b[39m run_method(\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mdriver_worker, method, args, kwargs)\n\u001b[1;32m     59\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m [answer]\n",
+      "File \u001b[0;32m~/sky_workdir/.venv/lib/python3.10/site-packages/vllm/utils/__init__.py:2985\u001b[0m, in \u001b[0;36mrun_method\u001b[0;34m()\u001b[0m\n\u001b[1;32m   2984\u001b[0m     func \u001b[38;5;241m=\u001b[39m partial(method, obj)  \u001b[38;5;66;03m# type: ignore\u001b[39;00m\n\u001b[0;32m-> 2985\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m func(\u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs)\n",
+      "File \u001b[0;32m~/sky_workdir/.venv/lib/python3.10/site-packages/vllm/worker/worker.py:211\u001b[0m, in \u001b[0;36mload_model\u001b[0;34m()\u001b[0m\n\u001b[1;32m    210\u001b[0m \u001b[38;5;28;01mwith\u001b[39;00m context:\n\u001b[0;32m--> 211\u001b[0m     \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mmodel_runner\u001b[38;5;241m.\u001b[39mload_model()\n",
+      "File \u001b[0;32m~/sky_workdir/.venv/lib/python3.10/site-packages/vllm/worker/model_runner.py:1086\u001b[0m, in \u001b[0;36mload_model\u001b[0;34m()\u001b[0m\n\u001b[1;32m   1085\u001b[0m time_before_load \u001b[38;5;241m=\u001b[39m time\u001b[38;5;241m.\u001b[39mperf_counter()\n\u001b[0;32m-> 1086\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mmodel \u001b[38;5;241m=\u001b[39m get_model(vllm_config\u001b[38;5;241m=\u001b[39m\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mvllm_config)\n\u001b[1;32m   1087\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mlora_config:\n",
+      "File \u001b[0;32m~/sky_workdir/.venv/lib/python3.10/site-packages/vllm/model_executor/model_loader/__init__.py:59\u001b[0m, in \u001b[0;36mget_model\u001b[0;34m()\u001b[0m\n\u001b[1;32m     58\u001b[0m     model_config \u001b[38;5;241m=\u001b[39m vllm_config\u001b[38;5;241m.\u001b[39mmodel_config\n\u001b[0;32m---> 59\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m loader\u001b[38;5;241m.\u001b[39mload_model(vllm_config\u001b[38;5;241m=\u001b[39mvllm_config,\n\u001b[1;32m     60\u001b[0m                          model_config\u001b[38;5;241m=\u001b[39mmodel_config)\n",
+      "File \u001b[0;32m~/sky_workdir/.venv/lib/python3.10/site-packages/vllm/model_executor/model_loader/base_loader.py:49\u001b[0m, in \u001b[0;36mload_model\u001b[0;34m()\u001b[0m\n\u001b[1;32m     48\u001b[0m \u001b[38;5;66;03m# Quantization does not happen in `load_weights` but after it\u001b[39;00m\n\u001b[0;32m---> 49\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mload_weights(model, model_config)\n\u001b[1;32m     50\u001b[0m process_weights_after_loading(model, model_config, target_device)\n",
+      "File \u001b[0;32m~/sky_workdir/.venv/lib/python3.10/site-packages/vllm/model_executor/model_loader/bitsandbytes_loader.py:731\u001b[0m, in \u001b[0;36mload_weights\u001b[0;34m()\u001b[0m\n\u001b[1;32m    730\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_verify_model_compatibility(model, model_config)\n\u001b[0;32m--> 731\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_initialize_loader_state(model, model_config)\n\u001b[1;32m    733\u001b[0m logger\u001b[38;5;241m.\u001b[39minfo(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mLoading weights with BitsAndBytes quantization. \u001b[39m\u001b[38;5;124m\"\u001b[39m\n\u001b[1;32m    734\u001b[0m             \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mMay take a while ...\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n",
+      "File \u001b[0;32m~/sky_workdir/.venv/lib/python3.10/site-packages/vllm/model_executor/model_loader/bitsandbytes_loader.py:524\u001b[0m, in \u001b[0;36m_initialize_loader_state\u001b[0;34m()\u001b[0m\n\u001b[1;32m    522\u001b[0m     \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mweight_mapper \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mlambda\u001b[39;00m name: hf_to_vllm_mapper\u001b[38;5;241m.\u001b[39m_map_name(name)\n\u001b[0;32m--> 524\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_get_bnb_target_modules(model)\n\u001b[1;32m    525\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_classify_module_sharding(model)\n",
+      "File \u001b[0;32m~/sky_workdir/.venv/lib/python3.10/site-packages/vllm/model_executor/model_loader/bitsandbytes_loader.py:441\u001b[0m, in \u001b[0;36m_get_bnb_target_modules\u001b[0;34m()\u001b[0m\n\u001b[1;32m    439\u001b[0m             \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mtarget_modules\u001b[38;5;241m.\u001b[39mappend(rep_name)\n\u001b[0;32m--> 441\u001b[0m \u001b[38;5;28;01massert\u001b[39;00m (\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mtarget_modules\n\u001b[1;32m    442\u001b[0m         ), \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mvLLM currently does not support BNB quantization for\u001b[39m\u001b[38;5;124m\"\u001b[39m\n\u001b[1;32m    443\u001b[0m \u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124m \u001b[39m\u001b[38;5;132;01m{\u001b[39;00m\u001b[38;5;28mtype\u001b[39m(model)\u001b[38;5;241m.\u001b[39m\u001b[38;5;18m__name__\u001b[39m\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m\"\u001b[39m\n",
+      "\u001b[0;31mAssertionError\u001b[0m: vLLM currently does not support BNB quantization for",
+      "\nDuring handling of the above exception, another exception occurred:\n",
+      "\u001b[0;31mRuntimeError\u001b[0m                              Traceback (most recent call last)",
+      "Cell \u001b[0;32mIn[3], line 16\u001b[0m\n\u001b[1;32m     10\u001b[0m backend \u001b[38;5;241m=\u001b[39m LocalBackend()\n\u001b[1;32m     11\u001b[0m model \u001b[38;5;241m=\u001b[39m art\u001b[38;5;241m.\u001b[39mTrainableModel(\n\u001b[1;32m     12\u001b[0m     name\u001b[38;5;241m=\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124m009\u001b[39m\u001b[38;5;124m\"\u001b[39m,\n\u001b[1;32m     13\u001b[0m     project\u001b[38;5;241m=\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124myes-no-maybe-vision\u001b[39m\u001b[38;5;124m\"\u001b[39m,\n\u001b[1;32m     14\u001b[0m     base_model\u001b[38;5;241m=\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mQwen/Qwen3-VL-8B-Instruct\u001b[39m\u001b[38;5;124m\"\u001b[39m,\n\u001b[1;32m     15\u001b[0m )\n\u001b[0;32m---> 16\u001b[0m \u001b[38;5;28;01mawait\u001b[39;00m model\u001b[38;5;241m.\u001b[39mregister(backend)\n\u001b[1;32m     19\u001b[0m \u001b[38;5;28;01masync\u001b[39;00m \u001b[38;5;28;01mdef\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[38;5;21mrollout\u001b[39m(client: openai\u001b[38;5;241m.\u001b[39mAsyncOpenAI, image_path: \u001b[38;5;28mstr\u001b[39m) \u001b[38;5;241m-\u001b[39m\u001b[38;5;241m>\u001b[39m art\u001b[38;5;241m.\u001b[39mTrajectory:\n\u001b[1;32m     20\u001b[0m     messages: art\u001b[38;5;241m.\u001b[39mMessages \u001b[38;5;241m=\u001b[39m [\n\u001b[1;32m     21\u001b[0m         {\n\u001b[1;32m     22\u001b[0m             \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mrole\u001b[39m\u001b[38;5;124m\"\u001b[39m: \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124muser\u001b[39m\u001b[38;5;124m\"\u001b[39m,\n\u001b[1;32m     23\u001b[0m             \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mcontent\u001b[39m\u001b[38;5;124m\"\u001b[39m: [{\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mtype\u001b[39m\u001b[38;5;124m\"\u001b[39m: \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mimage_url\u001b[39m\u001b[38;5;124m\"\u001b[39m, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mimage_url\u001b[39m\u001b[38;5;124m\"\u001b[39m: {\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124murl\u001b[39m\u001b[38;5;124m\"\u001b[39m: image_path}}],\n\u001b[1;32m     24\u001b[0m         }\n\u001b[1;32m     25\u001b[0m     ]\n",
+      "File \u001b[0;32m~/sky_workdir/src/art/model.py:335\u001b[0m, in \u001b[0;36mTrainableModel.register\u001b[0;34m(self, backend, _openai_client_config)\u001b[0m\n\u001b[1;32m    329\u001b[0m \u001b[38;5;28;01masync\u001b[39;00m \u001b[38;5;28;01mdef\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[38;5;21mregister\u001b[39m(\n\u001b[1;32m    330\u001b[0m     \u001b[38;5;28mself\u001b[39m,\n\u001b[1;32m    331\u001b[0m     backend: \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mBackend\u001b[39m\u001b[38;5;124m\"\u001b[39m,\n\u001b[1;32m    332\u001b[0m     _openai_client_config: dev\u001b[38;5;241m.\u001b[39mOpenAIServerConfig \u001b[38;5;241m|\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m,\n\u001b[1;32m    333\u001b[0m ) \u001b[38;5;241m-\u001b[39m\u001b[38;5;241m>\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m:\n\u001b[1;32m    334\u001b[0m     \u001b[38;5;28;01mawait\u001b[39;00m \u001b[38;5;28msuper\u001b[39m()\u001b[38;5;241m.\u001b[39mregister(backend)\n\u001b[0;32m--> 335\u001b[0m     base_url, api_key \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mawait\u001b[39;00m backend\u001b[38;5;241m.\u001b[39m_prepare_backend_for_training(\n\u001b[1;32m    336\u001b[0m         \u001b[38;5;28mself\u001b[39m, _openai_client_config\n\u001b[1;32m    337\u001b[0m     )\n\u001b[1;32m    339\u001b[0m     \u001b[38;5;66;03m# Populate the top-level inference fields so that the rest of the\u001b[39;00m\n\u001b[1;32m    340\u001b[0m     \u001b[38;5;66;03m# code (and any user code) can create an OpenAI client immediately.\u001b[39;00m\n\u001b[1;32m    341\u001b[0m     \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39minference_base_url \u001b[38;5;241m=\u001b[39m base_url\n",
+      "File \u001b[0;32m~/sky_workdir/src/art/local/backend.py:282\u001b[0m, in \u001b[0;36mLocalBackend._prepare_backend_for_training\u001b[0;34m(self, model, config)\u001b[0m\n\u001b[1;32m    276\u001b[0m \u001b[38;5;28;01masync\u001b[39;00m \u001b[38;5;28;01mdef\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[38;5;21m_prepare_backend_for_training\u001b[39m(\n\u001b[1;32m    277\u001b[0m     \u001b[38;5;28mself\u001b[39m,\n\u001b[1;32m    278\u001b[0m     model: TrainableModel,\n\u001b[1;32m    279\u001b[0m     config: dev\u001b[38;5;241m.\u001b[39mOpenAIServerConfig \u001b[38;5;241m|\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m,\n\u001b[1;32m    280\u001b[0m ) \u001b[38;5;241m-\u001b[39m\u001b[38;5;241m>\u001b[39m \u001b[38;5;28mtuple\u001b[39m[\u001b[38;5;28mstr\u001b[39m, \u001b[38;5;28mstr\u001b[39m]:\n\u001b[1;32m    281\u001b[0m     service \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mawait\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_get_service(model)\n\u001b[0;32m--> 282\u001b[0m     \u001b[38;5;28;01mawait\u001b[39;00m service\u001b[38;5;241m.\u001b[39mstart_openai_server(config\u001b[38;5;241m=\u001b[39mconfig)\n\u001b[1;32m    283\u001b[0m     server_args \u001b[38;5;241m=\u001b[39m (config \u001b[38;5;129;01mor\u001b[39;00m {})\u001b[38;5;241m.\u001b[39mget(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mserver_args\u001b[39m\u001b[38;5;124m\"\u001b[39m, {})\n\u001b[1;32m    285\u001b[0m     base_url \u001b[38;5;241m=\u001b[39m \u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mhttp://\u001b[39m\u001b[38;5;132;01m{\u001b[39;00mserver_args\u001b[38;5;241m.\u001b[39mget(\u001b[38;5;124m'\u001b[39m\u001b[38;5;124mhost\u001b[39m\u001b[38;5;124m'\u001b[39m,\u001b[38;5;250m \u001b[39m\u001b[38;5;124m'\u001b[39m\u001b[38;5;124m0.0.0.0\u001b[39m\u001b[38;5;124m'\u001b[39m)\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m:\u001b[39m\u001b[38;5;132;01m{\u001b[39;00mserver_args\u001b[38;5;241m.\u001b[39mget(\u001b[38;5;124m'\u001b[39m\u001b[38;5;124mport\u001b[39m\u001b[38;5;124m'\u001b[39m,\u001b[38;5;250m \u001b[39m\u001b[38;5;241m8000\u001b[39m)\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m/v1\u001b[39m\u001b[38;5;124m\"\u001b[39m\n",
+      "File \u001b[0;32m~/sky_workdir/src/mp_actors/traceback.py:26\u001b[0m, in \u001b[0;36mstreamline_tracebacks.<locals>.decorator.<locals>.async_wrapper\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m     24\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;01mawait\u001b[39;00m func(\u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs)\n\u001b[1;32m     25\u001b[0m \u001b[38;5;28;01mexcept\u001b[39;00m \u001b[38;5;167;01mException\u001b[39;00m \u001b[38;5;28;01mas\u001b[39;00m e:\n\u001b[0;32m---> 26\u001b[0m     \u001b[38;5;28;01mraise\u001b[39;00m e\u001b[38;5;241m.\u001b[39mwith_traceback(streamlined_traceback())\n",
+      "File \u001b[0;32m~/sky_workdir/src/art/unsloth/service.py:60\u001b[0m, in \u001b[0;36mstart_openai_server\u001b[0;34m()\u001b[0m\n\u001b[1;32m     58\u001b[0m     lora_path \u001b[38;5;241m=\u001b[39m get_step_checkpoint_dir(\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39moutput_dir, \u001b[38;5;241m0\u001b[39m)\n\u001b[1;32m     59\u001b[0m     os\u001b[38;5;241m.\u001b[39mmakedirs(os\u001b[38;5;241m.\u001b[39mpath\u001b[38;5;241m.\u001b[39mdirname(lora_path), exist_ok\u001b[38;5;241m=\u001b[39m\u001b[38;5;28;01mTrue\u001b[39;00m)\n\u001b[0;32m---> 60\u001b[0m     \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mstate\u001b[38;5;241m.\u001b[39mtrainer\u001b[38;5;241m.\u001b[39msave_model(lora_path)\n\u001b[1;32m     61\u001b[0m \u001b[38;5;28;01mawait\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mstop_openai_server()\n\u001b[1;32m     62\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_openai_server_task \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mawait\u001b[39;00m openai_server_task(\n\u001b[1;32m     63\u001b[0m     engine\u001b[38;5;241m=\u001b[39m\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mstate\u001b[38;5;241m.\u001b[39mvllm\u001b[38;5;241m.\u001b[39masync_engine,\n\u001b[1;32m     64\u001b[0m     config\u001b[38;5;241m=\u001b[39mdev\u001b[38;5;241m.\u001b[39mget_openai_server_config(\n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m     70\u001b[0m     ),\n\u001b[1;32m     71\u001b[0m )\n",
+      "File \u001b[0;32m~/miniconda3/lib/python3.10/functools.py:981\u001b[0m, in \u001b[0;36m__get__\u001b[0;34m()\u001b[0m\n\u001b[1;32m    979\u001b[0m val \u001b[38;5;241m=\u001b[39m cache\u001b[38;5;241m.\u001b[39mget(\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mattrname, _NOT_FOUND)\n\u001b[1;32m    980\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m val \u001b[38;5;129;01mis\u001b[39;00m _NOT_FOUND:\n\u001b[0;32m--> 981\u001b[0m     val \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mfunc(instance)\n\u001b[1;32m    982\u001b[0m     \u001b[38;5;28;01mtry\u001b[39;00m:\n\u001b[1;32m    983\u001b[0m         cache[\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mattrname] \u001b[38;5;241m=\u001b[39m val\n",
+      "File \u001b[0;32m~/sky_workdir/src/art/unsloth/service.py:45\u001b[0m, in \u001b[0;36mstate\u001b[0;34m()\u001b[0m\n\u001b[1;32m     41\u001b[0m \u001b[38;5;129m@functools\u001b[39m\u001b[38;5;241m.\u001b[39mcached_property\n\u001b[1;32m     42\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[38;5;21mstate\u001b[39m(\u001b[38;5;28mself\u001b[39m) \u001b[38;5;241m-\u001b[39m\u001b[38;5;241m>\u001b[39m \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mModelState\u001b[39m\u001b[38;5;124m\"\u001b[39m:\n\u001b[1;32m     43\u001b[0m     \u001b[38;5;28;01mfrom\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[38;5;21;01m.\u001b[39;00m\u001b[38;5;21;01mstate\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[38;5;28;01mimport\u001b[39;00m ModelState\n\u001b[0;32m---> 45\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m ModelState(\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mconfig)\n",
+      "File \u001b[0;32m~/sky_workdir/src/art/unsloth/state.py:82\u001b[0m, in \u001b[0;36m__init__\u001b[0;34m()\u001b[0m\n\u001b[1;32m     74\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m from_engine_args(\n\u001b[1;32m     75\u001b[0m         replace(engine_args, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mconfig\u001b[38;5;241m.\u001b[39mget(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mengine_args\u001b[39m\u001b[38;5;124m\"\u001b[39m, {})), \u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs\n\u001b[1;32m     76\u001b[0m     )\n\u001b[1;32m     78\u001b[0m AsyncLLMEngine\u001b[38;5;241m.\u001b[39mfrom_engine_args \u001b[38;5;241m=\u001b[39m _from_engine_args\n\u001b[1;32m     80\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mmodel, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mtokenizer \u001b[38;5;241m=\u001b[39m cast(\n\u001b[1;32m     81\u001b[0m     \u001b[38;5;28mtuple\u001b[39m[CausallLM, PreTrainedTokenizerBase],\n\u001b[0;32m---> 82\u001b[0m     unsloth\u001b[38;5;241m.\u001b[39mFastLanguageModel\u001b[38;5;241m.\u001b[39mfrom_pretrained(\u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mconfig\u001b[38;5;241m.\u001b[39mget(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124minit_args\u001b[39m\u001b[38;5;124m\"\u001b[39m, {})),\n\u001b[1;32m     83\u001b[0m )\n\u001b[1;32m     84\u001b[0m AsyncLLMEngine\u001b[38;5;241m.\u001b[39mfrom_engine_args \u001b[38;5;241m=\u001b[39m from_engine_args\n\u001b[1;32m     85\u001b[0m torch\u001b[38;5;241m.\u001b[39mcuda\u001b[38;5;241m.\u001b[39mempty_cache \u001b[38;5;241m=\u001b[39m empty_cache\n",
+      "File \u001b[0;32m~/sky_workdir/.venv/lib/python3.10/site-packages/unsloth/models/loader.py:431\u001b[0m, in \u001b[0;36mfrom_pretrained\u001b[0;34m()\u001b[0m\n\u001b[1;32m    414\u001b[0m     dispatch_model \u001b[38;5;241m=\u001b[39m FastQwen3Model \u001b[38;5;28;01mif\u001b[39;00m model_type \u001b[38;5;241m==\u001b[39m \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mqwen3\u001b[39m\u001b[38;5;124m\"\u001b[39m \u001b[38;5;28;01melse\u001b[39;00m FastQwen3MoeModel\n\u001b[1;32m    415\u001b[0m \u001b[38;5;66;03m# elif model_type == \"falcon_h1\":\u001b[39;00m\n\u001b[1;32m    416\u001b[0m \u001b[38;5;66;03m#     dispatch_model = FastFalconH1Model\u001b[39;00m\n\u001b[1;32m    417\u001b[0m \u001b[38;5;66;03m#     if not SUPPORTS_FALCON_H1:\u001b[39;00m\n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m    429\u001b[0m \u001b[38;5;66;03m#     dispatch_model = FastGraniteModel\u001b[39;00m\n\u001b[1;32m    430\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m--> 431\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m FastModel\u001b[38;5;241m.\u001b[39mfrom_pretrained(\n\u001b[1;32m    432\u001b[0m         model_name                 \u001b[38;5;241m=\u001b[39m old_model_name,\n\u001b[1;32m    433\u001b[0m         max_seq_length             \u001b[38;5;241m=\u001b[39m max_seq_length,\n\u001b[1;32m    434\u001b[0m         dtype                      \u001b[38;5;241m=\u001b[39m dtype,\n\u001b[1;32m    435\u001b[0m         load_in_4bit               \u001b[38;5;241m=\u001b[39m load_in_4bit,\n\u001b[1;32m    436\u001b[0m         load_in_8bit               \u001b[38;5;241m=\u001b[39m load_in_8bit,\n\u001b[1;32m    437\u001b[0m         load_in_16bit              \u001b[38;5;241m=\u001b[39m load_in_16bit,\n\u001b[1;32m    438\u001b[0m         full_finetuning            \u001b[38;5;241m=\u001b[39m full_finetuning,\n\u001b[1;32m    439\u001b[0m         token                      \u001b[38;5;241m=\u001b[39m token,\n\u001b[1;32m    440\u001b[0m         device_map                 \u001b[38;5;241m=\u001b[39m device_map,\n\u001b[1;32m    441\u001b[0m         rope_scaling               \u001b[38;5;241m=\u001b[39m rope_scaling, \u001b[38;5;66;03m# [TODO] No effect\u001b[39;00m\n\u001b[1;32m    442\u001b[0m         fix_tokenizer              \u001b[38;5;241m=\u001b[39m fix_tokenizer, \u001b[38;5;66;03m# [TODO] No effect\u001b[39;00m\n\u001b[1;32m    443\u001b[0m         trust_remote_code          \u001b[38;5;241m=\u001b[39m trust_remote_code,\n\u001b[1;32m    444\u001b[0m         use_gradient_checkpointing \u001b[38;5;241m=\u001b[39m use_gradient_checkpointing,\n\u001b[1;32m    445\u001b[0m         resize_model_vocab         \u001b[38;5;241m=\u001b[39m resize_model_vocab, \u001b[38;5;66;03m# [TODO] No effect\u001b[39;00m\n\u001b[1;32m    446\u001b[0m         revision                   \u001b[38;5;241m=\u001b[39m revision,\n\u001b[1;32m    447\u001b[0m         return_logits              \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mFalse\u001b[39;00m, \u001b[38;5;66;03m# Return logits\u001b[39;00m\n\u001b[1;32m    448\u001b[0m         fullgraph                  \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mTrue\u001b[39;00m, \u001b[38;5;66;03m# No graph breaks\u001b[39;00m\n\u001b[1;32m    449\u001b[0m         use_exact_model_name       \u001b[38;5;241m=\u001b[39m use_exact_model_name,\n\u001b[1;32m    450\u001b[0m         offload_embedding          \u001b[38;5;241m=\u001b[39m offload_embedding,\n\u001b[1;32m    451\u001b[0m \n\u001b[1;32m    452\u001b[0m         \u001b[38;5;66;03m# Pass vLLM/inference parameters\u001b[39;00m\n\u001b[1;32m    453\u001b[0m         fast_inference             \u001b[38;5;241m=\u001b[39m fast_inference,\n\u001b[1;32m    454\u001b[0m         gpu_memory_utilization     \u001b[38;5;241m=\u001b[39m gpu_memory_utilization,\n\u001b[1;32m    455\u001b[0m         float8_kv_cache            \u001b[38;5;241m=\u001b[39m float8_kv_cache,\n\u001b[1;32m    456\u001b[0m         random_state               \u001b[38;5;241m=\u001b[39m random_state,\n\u001b[1;32m    457\u001b[0m         max_lora_rank              \u001b[38;5;241m=\u001b[39m max_lora_rank,\n\u001b[1;32m    458\u001b[0m         disable_log_stats          \u001b[38;5;241m=\u001b[39m disable_log_stats,\n\u001b[1;32m    459\u001b[0m \n\u001b[1;32m    460\u001b[0m         \u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs,\n\u001b[1;32m    461\u001b[0m     )\n\u001b[1;32m    462\u001b[0m \u001b[38;5;28;01mpass\u001b[39;00m\n\u001b[1;32m    464\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m use_gradient_checkpointing \u001b[38;5;241m==\u001b[39m \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124munsloth\u001b[39m\u001b[38;5;124m\"\u001b[39m:\n",
+      "File \u001b[0;32m~/sky_workdir/.venv/lib/python3.10/site-packages/unsloth/models/loader.py:988\u001b[0m, in \u001b[0;36mfrom_pretrained\u001b[0;34m()\u001b[0m\n\u001b[1;32m    985\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m auto_model \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m:\n\u001b[1;32m    986\u001b[0m     auto_model \u001b[38;5;241m=\u001b[39m AutoModelForVision2Seq \u001b[38;5;28;01mif\u001b[39;00m is_vlm \u001b[38;5;28;01melse\u001b[39;00m AutoModelForCausalLM\n\u001b[0;32m--> 988\u001b[0m model, tokenizer \u001b[38;5;241m=\u001b[39m FastBaseModel\u001b[38;5;241m.\u001b[39mfrom_pretrained(\n\u001b[1;32m    989\u001b[0m     model_name        \u001b[38;5;241m=\u001b[39m model_name,\n\u001b[1;32m    990\u001b[0m     max_seq_length    \u001b[38;5;241m=\u001b[39m max_seq_length,\n\u001b[1;32m    991\u001b[0m     dtype             \u001b[38;5;241m=\u001b[39m _get_dtype(dtype),\n\u001b[1;32m    992\u001b[0m     load_in_4bit      \u001b[38;5;241m=\u001b[39m load_in_4bit,\n\u001b[1;32m    993\u001b[0m     load_in_8bit      \u001b[38;5;241m=\u001b[39m load_in_8bit,\n\u001b[1;32m    994\u001b[0m     load_in_16bit     \u001b[38;5;241m=\u001b[39m load_in_16bit,\n\u001b[1;32m    995\u001b[0m     full_finetuning   \u001b[38;5;241m=\u001b[39m full_finetuning,\n\u001b[1;32m    996\u001b[0m     token             \u001b[38;5;241m=\u001b[39m token,\n\u001b[1;32m    997\u001b[0m     device_map        \u001b[38;5;241m=\u001b[39m device_map,\n\u001b[1;32m    998\u001b[0m     trust_remote_code \u001b[38;5;241m=\u001b[39m trust_remote_code,\n\u001b[1;32m    999\u001b[0m     revision          \u001b[38;5;241m=\u001b[39m revision \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m is_peft \u001b[38;5;28;01melse\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m,\n\u001b[1;32m   1000\u001b[0m     model_types       \u001b[38;5;241m=\u001b[39m model_types,\n\u001b[1;32m   1001\u001b[0m     tokenizer_name    \u001b[38;5;241m=\u001b[39m tokenizer_name,\n\u001b[1;32m   1002\u001b[0m     auto_model        \u001b[38;5;241m=\u001b[39m auto_model,\n\u001b[1;32m   1003\u001b[0m     use_gradient_checkpointing \u001b[38;5;241m=\u001b[39m use_gradient_checkpointing,\n\u001b[1;32m   1004\u001b[0m     supports_sdpa     \u001b[38;5;241m=\u001b[39m supports_sdpa,\n\u001b[1;32m   1005\u001b[0m     whisper_language  \u001b[38;5;241m=\u001b[39m whisper_language,\n\u001b[1;32m   1006\u001b[0m     whisper_task      \u001b[38;5;241m=\u001b[39m whisper_task,\n\u001b[1;32m   1007\u001b[0m     auto_config       \u001b[38;5;241m=\u001b[39m model_config,\n\u001b[1;32m   1008\u001b[0m     offload_embedding \u001b[38;5;241m=\u001b[39m offload_embedding,\n\u001b[1;32m   1009\u001b[0m \n\u001b[1;32m   1010\u001b[0m     \u001b[38;5;66;03m# Pass vLLM/inference parameters\u001b[39;00m\n\u001b[1;32m   1011\u001b[0m     fast_inference         \u001b[38;5;241m=\u001b[39m fast_inference,\n\u001b[1;32m   1012\u001b[0m     gpu_memory_utilization \u001b[38;5;241m=\u001b[39m gpu_memory_utilization,\n\u001b[1;32m   1013\u001b[0m     float8_kv_cache        \u001b[38;5;241m=\u001b[39m float8_kv_cache,\n\u001b[1;32m   1014\u001b[0m     random_state           \u001b[38;5;241m=\u001b[39m random_state,\n\u001b[1;32m   1015\u001b[0m     max_lora_rank          \u001b[38;5;241m=\u001b[39m max_lora_rank,\n\u001b[1;32m   1016\u001b[0m     disable_log_stats      \u001b[38;5;241m=\u001b[39m disable_log_stats,\n\u001b[1;32m   1017\u001b[0m \n\u001b[1;32m   1018\u001b[0m     \u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs,\n\u001b[1;32m   1019\u001b[0m )\n\u001b[1;32m   1021\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m resize_model_vocab \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m:\n\u001b[1;32m   1022\u001b[0m     model\u001b[38;5;241m.\u001b[39mresize_token_embeddings(resize_model_vocab)\n",
+      "File \u001b[0;32m~/sky_workdir/.venv/lib/python3.10/site-packages/unsloth/models/vision.py:628\u001b[0m, in \u001b[0;36mfrom_pretrained\u001b[0;34m()\u001b[0m\n\u001b[1;32m    625\u001b[0m \u001b[38;5;28;01mpass\u001b[39;00m\n\u001b[1;32m    627\u001b[0m \u001b[38;5;66;03m# Load vLLM first\u001b[39;00m\n\u001b[0;32m--> 628\u001b[0m llm \u001b[38;5;241m=\u001b[39m load_vllm(\u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mload_vllm_kwargs)\n\u001b[1;32m    630\u001b[0m \u001b[38;5;66;03m# Convert to HF format\u001b[39;00m\n\u001b[1;32m    631\u001b[0m _, quant_state_dict \u001b[38;5;241m=\u001b[39m get_vllm_state_dict(\n\u001b[1;32m    632\u001b[0m     llm,\n\u001b[1;32m    633\u001b[0m     config \u001b[38;5;241m=\u001b[39m model_config,\n\u001b[1;32m    634\u001b[0m     is_vision_model \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mTrue\u001b[39;00m,\n\u001b[1;32m    635\u001b[0m )\n",
+      "File \u001b[0;32m~/sky_workdir/.venv/lib/python3.10/site-packages/unsloth_zoo/vllm_utils.py:1792\u001b[0m, in \u001b[0;36mload_vllm\u001b[0;34m()\u001b[0m\n\u001b[1;32m   1787\u001b[0m             \u001b[38;5;28mprint\u001b[39m(\n\u001b[1;32m   1788\u001b[0m                 \u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mUnsloth: Retrying vLLM to process \u001b[39m\u001b[38;5;132;01m{\u001b[39;00mapprox_max_num_seqs\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m sequences and \u001b[39m\u001b[38;5;132;01m{\u001b[39;00mmax_num_batched_tokens\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m tokens in tandem.\u001b[39m\u001b[38;5;130;01m\\n\u001b[39;00m\u001b[38;5;124m\"\u001b[39m\\\n\u001b[1;32m   1789\u001b[0m                 \u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mError:\u001b[39m\u001b[38;5;130;01m\\n\u001b[39;00m\u001b[38;5;132;01m{\u001b[39;00merror\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m\"\u001b[39m\n\u001b[1;32m   1790\u001b[0m             )\n\u001b[1;32m   1791\u001b[0m         \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m-> 1792\u001b[0m             \u001b[38;5;28;01mraise\u001b[39;00m \u001b[38;5;167;01mRuntimeError\u001b[39;00m(error)\n\u001b[1;32m   1793\u001b[0m     \u001b[38;5;28;01mpass\u001b[39;00m\n\u001b[1;32m   1794\u001b[0m \u001b[38;5;28;01mpass\u001b[39;00m\n",
+      "\u001b[0;31mRuntimeError\u001b[0m: vLLM currently does not support BNB quantization for"
+     ]
+    }
+   ],
    "source": [
     "import openai\n",
     "from dotenv import load_dotenv\n",
@@ -47,7 +266,7 @@
     "model = art.TrainableModel(\n",
     "    name=\"009\",\n",
     "    project=\"yes-no-maybe-vision\",\n",
-    "    base_model=\"Qwen/Qwen2.5-VL-7B-Instruct\",\n",
+    "    base_model=\"Qwen/Qwen3-VL-8B-Instruct\",\n",
     ")\n",
     "await model.register(backend)\n",
     "\n",
diff --git a/pyproject.toml b/pyproject.toml
index 5179dfdf..de1c867b 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -18,8 +18,8 @@ backend = [
     "peft>=0.14.0",
     "hf-xet>=1.1.0",
     "bitsandbytes>=0.45.2",
-    "unsloth==2025.10.3",
-    "unsloth-zoo==2025.10.3",
+    "unsloth==2025.10.8",
+    "unsloth-zoo==2025.10.9",
     "vllm>=0.9.2,<=0.10.0",
     "torchtune",
     "trl>=0.19.0",
@@ -32,7 +32,7 @@ backend = [
     "setuptools>=78.1.0",
     "wandb==0.21.0",
     "polars>=1.26.0",
-    "transformers==4.53.2",
+    "transformers==4.56.2",
     "trl==0.20.0",
     "nbclient>=0.10.1",
     "pytest>=8.4.1",
diff --git a/uv.lock b/uv.lock
index 1a20b540..7db56dfc 100644
--- a/uv.lock
+++ b/uv.lock
@@ -4235,12 +4235,12 @@ requires-dist = [
     { name = "torch", marker = "extra == 'backend'", specifier = ">=2.7.0" },
     { name = "torchao", marker = "extra == 'backend'", specifier = ">=0.9.0" },
     { name = "torchtune", marker = "extra == 'backend'", git = "https://github.com/pytorch/torchtune.git?rev=2344509cf83bd886538fe3e8263e5145d1afb5c2" },
-    { name = "transformers", marker = "extra == 'backend'", specifier = "==4.53.2" },
+    { name = "transformers", marker = "extra == 'backend'", specifier = "==4.56.2" },
     { name = "trl", marker = "extra == 'backend'", specifier = "==0.20.0" },
     { name = "trl", marker = "extra == 'backend'", specifier = ">=0.19.0" },
     { name = "typer", specifier = ">=0.15.2" },
-    { name = "unsloth", marker = "extra == 'backend'", specifier = "==2025.10.3" },
-    { name = "unsloth-zoo", marker = "extra == 'backend'", specifier = "==2025.10.3" },
+    { name = "unsloth", marker = "extra == 'backend'", specifier = "==2025.10.8" },
+    { name = "unsloth-zoo", marker = "extra == 'backend'", specifier = "==2025.10.9" },
     { name = "vllm", marker = "extra == 'backend'", specifier = ">=0.9.2,<=0.10.0" },
     { name = "wandb", marker = "extra == 'backend'", specifier = "==0.21.0" },
     { name = "weave", specifier = ">=0.51.51" },
@@ -7128,27 +7128,27 @@ wheels = [
 
 [[package]]
 name = "tokenizers"
-version = "0.21.4"
+version = "0.22.1"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "huggingface-hub" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/c2/2f/402986d0823f8d7ca139d969af2917fefaa9b947d1fb32f6168c509f2492/tokenizers-0.21.4.tar.gz", hash = "sha256:fa23f85fbc9a02ec5c6978da172cdcbac23498c3ca9f3645c5c68740ac007880", size = 351253, upload-time = "2025-07-28T15:48:54.325Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/1c/46/fb6854cec3278fbfa4a75b50232c77622bc517ac886156e6afbfa4d8fc6e/tokenizers-0.22.1.tar.gz", hash = "sha256:61de6522785310a309b3407bac22d99c4db5dba349935e99e4d15ea2226af2d9", size = 363123, upload-time = "2025-09-19T09:49:23.424Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/98/c6/fdb6f72bf6454f52eb4a2510be7fb0f614e541a2554d6210e370d85efff4/tokenizers-0.21.4-cp39-abi3-macosx_10_12_x86_64.whl", hash = "sha256:2ccc10a7c3bcefe0f242867dc914fc1226ee44321eb618cfe3019b5df3400133", size = 2863987, upload-time = "2025-07-28T15:48:44.877Z" },
-    { url = "https://files.pythonhosted.org/packages/8d/a6/28975479e35ddc751dc1ddc97b9b69bf7fcf074db31548aab37f8116674c/tokenizers-0.21.4-cp39-abi3-macosx_11_0_arm64.whl", hash = "sha256:5e2f601a8e0cd5be5cc7506b20a79112370b9b3e9cb5f13f68ab11acd6ca7d60", size = 2732457, upload-time = "2025-07-28T15:48:43.265Z" },
-    { url = "https://files.pythonhosted.org/packages/aa/8f/24f39d7b5c726b7b0be95dca04f344df278a3fe3a4deb15a975d194cbb32/tokenizers-0.21.4-cp39-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:39b376f5a1aee67b4d29032ee85511bbd1b99007ec735f7f35c8a2eb104eade5", size = 3012624, upload-time = "2025-07-28T13:22:43.895Z" },
-    { url = "https://files.pythonhosted.org/packages/58/47/26358925717687a58cb74d7a508de96649544fad5778f0cd9827398dc499/tokenizers-0.21.4-cp39-abi3-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:2107ad649e2cda4488d41dfd031469e9da3fcbfd6183e74e4958fa729ffbf9c6", size = 2939681, upload-time = "2025-07-28T13:22:47.499Z" },
-    { url = "https://files.pythonhosted.org/packages/99/6f/cc300fea5db2ab5ddc2c8aea5757a27b89c84469899710c3aeddc1d39801/tokenizers-0.21.4-cp39-abi3-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:3c73012da95afafdf235ba80047699df4384fdc481527448a078ffd00e45a7d9", size = 3247445, upload-time = "2025-07-28T15:48:39.711Z" },
-    { url = "https://files.pythonhosted.org/packages/be/bf/98cb4b9c3c4afd8be89cfa6423704337dc20b73eb4180397a6e0d456c334/tokenizers-0.21.4-cp39-abi3-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:f23186c40395fc390d27f519679a58023f368a0aad234af145e0f39ad1212732", size = 3428014, upload-time = "2025-07-28T13:22:49.569Z" },
-    { url = "https://files.pythonhosted.org/packages/75/c7/96c1cc780e6ca7f01a57c13235dd05b7bc1c0f3588512ebe9d1331b5f5ae/tokenizers-0.21.4-cp39-abi3-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:cc88bb34e23a54cc42713d6d98af5f1bf79c07653d24fe984d2d695ba2c922a2", size = 3193197, upload-time = "2025-07-28T13:22:51.471Z" },
-    { url = "https://files.pythonhosted.org/packages/f2/90/273b6c7ec78af547694eddeea9e05de771278bd20476525ab930cecaf7d8/tokenizers-0.21.4-cp39-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:51b7eabb104f46c1c50b486520555715457ae833d5aee9ff6ae853d1130506ff", size = 3115426, upload-time = "2025-07-28T15:48:41.439Z" },
-    { url = "https://files.pythonhosted.org/packages/91/43/c640d5a07e95f1cf9d2c92501f20a25f179ac53a4f71e1489a3dcfcc67ee/tokenizers-0.21.4-cp39-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:714b05b2e1af1288bd1bc56ce496c4cebb64a20d158ee802887757791191e6e2", size = 9089127, upload-time = "2025-07-28T15:48:46.472Z" },
-    { url = "https://files.pythonhosted.org/packages/44/a1/dd23edd6271d4dca788e5200a807b49ec3e6987815cd9d0a07ad9c96c7c2/tokenizers-0.21.4-cp39-abi3-musllinux_1_2_armv7l.whl", hash = "sha256:1340ff877ceedfa937544b7d79f5b7becf33a4cfb58f89b3b49927004ef66f78", size = 9055243, upload-time = "2025-07-28T15:48:48.539Z" },
-    { url = "https://files.pythonhosted.org/packages/21/2b/b410d6e9021c4b7ddb57248304dc817c4d4970b73b6ee343674914701197/tokenizers-0.21.4-cp39-abi3-musllinux_1_2_i686.whl", hash = "sha256:3c1f4317576e465ac9ef0d165b247825a2a4078bcd01cba6b54b867bdf9fdd8b", size = 9298237, upload-time = "2025-07-28T15:48:50.443Z" },
-    { url = "https://files.pythonhosted.org/packages/b7/0a/42348c995c67e2e6e5c89ffb9cfd68507cbaeb84ff39c49ee6e0a6dd0fd2/tokenizers-0.21.4-cp39-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:c212aa4e45ec0bb5274b16b6f31dd3f1c41944025c2358faaa5782c754e84c24", size = 9461980, upload-time = "2025-07-28T15:48:52.325Z" },
-    { url = "https://files.pythonhosted.org/packages/3d/d3/dacccd834404cd71b5c334882f3ba40331ad2120e69ded32cf5fda9a7436/tokenizers-0.21.4-cp39-abi3-win32.whl", hash = "sha256:6c42a930bc5f4c47f4ea775c91de47d27910881902b0f20e4990ebe045a415d0", size = 2329871, upload-time = "2025-07-28T15:48:56.841Z" },
-    { url = "https://files.pythonhosted.org/packages/41/f2/fd673d979185f5dcbac4be7d09461cbb99751554ffb6718d0013af8604cb/tokenizers-0.21.4-cp39-abi3-win_amd64.whl", hash = "sha256:475d807a5c3eb72c59ad9b5fcdb254f6e17f53dfcbb9903233b0dfa9c943b597", size = 2507568, upload-time = "2025-07-28T15:48:55.456Z" },
+    { url = "https://files.pythonhosted.org/packages/bf/33/f4b2d94ada7ab297328fc671fed209368ddb82f965ec2224eb1892674c3a/tokenizers-0.22.1-cp39-abi3-macosx_10_12_x86_64.whl", hash = "sha256:59fdb013df17455e5f950b4b834a7b3ee2e0271e6378ccb33aa74d178b513c73", size = 3069318, upload-time = "2025-09-19T09:49:11.848Z" },
+    { url = "https://files.pythonhosted.org/packages/1c/58/2aa8c874d02b974990e89ff95826a4852a8b2a273c7d1b4411cdd45a4565/tokenizers-0.22.1-cp39-abi3-macosx_11_0_arm64.whl", hash = "sha256:8d4e484f7b0827021ac5f9f71d4794aaef62b979ab7608593da22b1d2e3c4edc", size = 2926478, upload-time = "2025-09-19T09:49:09.759Z" },
+    { url = "https://files.pythonhosted.org/packages/1e/3b/55e64befa1e7bfea963cf4b787b2cea1011362c4193f5477047532ce127e/tokenizers-0.22.1-cp39-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:19d2962dd28bc67c1f205ab180578a78eef89ac60ca7ef7cbe9635a46a56422a", size = 3256994, upload-time = "2025-09-19T09:48:56.701Z" },
+    { url = "https://files.pythonhosted.org/packages/71/0b/fbfecf42f67d9b7b80fde4aabb2b3110a97fac6585c9470b5bff103a80cb/tokenizers-0.22.1-cp39-abi3-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:38201f15cdb1f8a6843e6563e6e79f4abd053394992b9bbdf5213ea3469b4ae7", size = 3153141, upload-time = "2025-09-19T09:48:59.749Z" },
+    { url = "https://files.pythonhosted.org/packages/17/a9/b38f4e74e0817af8f8ef925507c63c6ae8171e3c4cb2d5d4624bf58fca69/tokenizers-0.22.1-cp39-abi3-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:d1cbe5454c9a15df1b3443c726063d930c16f047a3cc724b9e6e1a91140e5a21", size = 3508049, upload-time = "2025-09-19T09:49:05.868Z" },
+    { url = "https://files.pythonhosted.org/packages/d2/48/dd2b3dac46bb9134a88e35d72e1aa4869579eacc1a27238f1577270773ff/tokenizers-0.22.1-cp39-abi3-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:e7d094ae6312d69cc2a872b54b91b309f4f6fbce871ef28eb27b52a98e4d0214", size = 3710730, upload-time = "2025-09-19T09:49:01.832Z" },
+    { url = "https://files.pythonhosted.org/packages/93/0e/ccabc8d16ae4ba84a55d41345207c1e2ea88784651a5a487547d80851398/tokenizers-0.22.1-cp39-abi3-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:afd7594a56656ace95cdd6df4cca2e4059d294c5cfb1679c57824b605556cb2f", size = 3412560, upload-time = "2025-09-19T09:49:03.867Z" },
+    { url = "https://files.pythonhosted.org/packages/d0/c6/dc3a0db5a6766416c32c034286d7c2d406da1f498e4de04ab1b8959edd00/tokenizers-0.22.1-cp39-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:e2ef6063d7a84994129732b47e7915e8710f27f99f3a3260b8a38fc7ccd083f4", size = 3250221, upload-time = "2025-09-19T09:49:07.664Z" },
+    { url = "https://files.pythonhosted.org/packages/d7/a6/2c8486eef79671601ff57b093889a345dd3d576713ef047776015dc66de7/tokenizers-0.22.1-cp39-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:ba0a64f450b9ef412c98f6bcd2a50c6df6e2443b560024a09fa6a03189726879", size = 9345569, upload-time = "2025-09-19T09:49:14.214Z" },
+    { url = "https://files.pythonhosted.org/packages/6b/16/32ce667f14c35537f5f605fe9bea3e415ea1b0a646389d2295ec348d5657/tokenizers-0.22.1-cp39-abi3-musllinux_1_2_armv7l.whl", hash = "sha256:331d6d149fa9c7d632cde4490fb8bbb12337fa3a0232e77892be656464f4b446", size = 9271599, upload-time = "2025-09-19T09:49:16.639Z" },
+    { url = "https://files.pythonhosted.org/packages/51/7c/a5f7898a3f6baa3fc2685c705e04c98c1094c523051c805cdd9306b8f87e/tokenizers-0.22.1-cp39-abi3-musllinux_1_2_i686.whl", hash = "sha256:607989f2ea68a46cb1dfbaf3e3aabdf3f21d8748312dbeb6263d1b3b66c5010a", size = 9533862, upload-time = "2025-09-19T09:49:19.146Z" },
+    { url = "https://files.pythonhosted.org/packages/36/65/7e75caea90bc73c1dd8d40438adf1a7bc26af3b8d0a6705ea190462506e1/tokenizers-0.22.1-cp39-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:a0f307d490295717726598ef6fa4f24af9d484809223bbc253b201c740a06390", size = 9681250, upload-time = "2025-09-19T09:49:21.501Z" },
+    { url = "https://files.pythonhosted.org/packages/30/2c/959dddef581b46e6209da82df3b78471e96260e2bc463f89d23b1bf0e52a/tokenizers-0.22.1-cp39-abi3-win32.whl", hash = "sha256:b5120eed1442765cd90b903bb6cfef781fd8fe64e34ccaecbae4c619b7b12a82", size = 2472003, upload-time = "2025-09-19T09:49:27.089Z" },
+    { url = "https://files.pythonhosted.org/packages/b3/46/e33a8c93907b631a99377ef4c5f817ab453d0b34f93529421f42ff559671/tokenizers-0.22.1-cp39-abi3-win_amd64.whl", hash = "sha256:65fd6e3fb11ca1e78a6a93602490f134d1fdeb13bcef99389d5102ea318ed138", size = 2674684, upload-time = "2025-09-19T09:49:24.953Z" },
 ]
 
 [[package]]
@@ -7262,11 +7262,11 @@ wheels = [
 
 [[package]]
 name = "torchao"
-version = "0.12.0"
+version = "0.14.0"
 source = { registry = "https://pypi.org/simple" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/72/5e/f5df5e0bbc7d84e5da3b4599b5bad58f4a6657b22bcae64dd741faee80ab/torchao-0.12.0-cp39-abi3-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:326ea2800cc7d9d50f0d17742ad923e5c6d4c4dd5942558f4ed13db00bdebc7c", size = 6777967, upload-time = "2025-07-17T17:50:13.567Z" },
-    { url = "https://files.pythonhosted.org/packages/6c/5f/6bf9b5bed6d31e286516d23e1db7320d2ccfbf1b2234749833ad1e3d25a5/torchao-0.12.0-py3-none-any.whl", hash = "sha256:103f2a9164d2e4f705332af1aafbb8473eadd14d9164e45857ca187cde1f13d2", size = 962232, upload-time = "2025-07-17T17:50:15.119Z" },
+    { url = "https://files.pythonhosted.org/packages/28/5a/72d932efcff037fea3466b0e0c9a82d280e9083da10fe79af8505ac71348/torchao-0.14.0-cp310-abi3-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:3508151beb43ab6dd0f70427389f387cf1cd367f2326e9c6102133991e147b67", size = 7197166, upload-time = "2025-10-13T21:16:39.029Z" },
+    { url = "https://files.pythonhosted.org/packages/6f/f0/975479131e6d8365854b97334688dd156a6d67824cc24b65a6e661e93800/torchao-0.14.0-py3-none-any.whl", hash = "sha256:c5a939e035f8cc0493f2bb18a3d01c1db8139662f7fcf49a6df84366218aa95e", size = 1067582, upload-time = "2025-10-13T21:16:40.985Z" },
 ]
 
 [[package]]
@@ -7419,7 +7419,7 @@ wheels = [
 
 [[package]]
 name = "transformers"
-version = "4.53.2"
+version = "4.56.2"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "filelock" },
@@ -7433,9 +7433,9 @@ dependencies = [
     { name = "tokenizers" },
     { name = "tqdm" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/4c/67/80f51466ec447028fd84469b208eb742533ce06cc8fad2e3181380199e5c/transformers-4.53.2.tar.gz", hash = "sha256:6c3ed95edfb1cba71c4245758f1b4878c93bf8cde77d076307dacb2cbbd72be2", size = 9201233, upload-time = "2025-07-11T12:39:08.742Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/e5/82/0bcfddd134cdf53440becb5e738257cc3cf34cf229d63b57bfd288e6579f/transformers-4.56.2.tar.gz", hash = "sha256:5e7c623e2d7494105c726dd10f6f90c2c99a55ebe86eef7233765abd0cb1c529", size = 9844296, upload-time = "2025-09-19T15:16:26.778Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/96/88/beb33a79a382fcd2aed0be5222bdc47f41e4bfe7aaa90ae1374f1d8ea2af/transformers-4.53.2-py3-none-any.whl", hash = "sha256:db8f4819bb34f000029c73c3c557e7d06fc1b8e612ec142eecdae3947a9c78bf", size = 10826609, upload-time = "2025-07-11T12:39:05.461Z" },
+    { url = "https://files.pythonhosted.org/packages/70/26/2591b48412bde75e33bfd292034103ffe41743cacd03120e3242516cd143/transformers-4.56.2-py3-none-any.whl", hash = "sha256:79c03d0e85b26cb573c109ff9eafa96f3c8d4febfd8a0774e8bba32702dd6dde", size = 11608055, upload-time = "2025-09-19T15:16:23.736Z" },
 ]
 
 [[package]]
@@ -7658,7 +7658,7 @@ wheels = [
 
 [[package]]
 name = "unsloth"
-version = "2025.10.3"
+version = "2025.10.8"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "accelerate" },
@@ -7685,14 +7685,14 @@ dependencies = [
     { name = "wheel" },
     { name = "xformers", marker = "(platform_machine == 'AMD64' and 'linux' in sys_platform) or (platform_machine == 'x86_64' and 'linux' in sys_platform) or (platform_machine == 'AMD64' and sys_platform == 'win32') or (platform_machine == 'x86_64' and sys_platform == 'win32')" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/f9/f2/6ddf93ebbbbc7e8f2dca705322a0fbb36cf55434669dffa922d69f432cb4/unsloth-2025.10.3.tar.gz", hash = "sha256:eb60921af0b78bb102f5856cfc52b6a7cfa984ba4470a596de2695065691987a", size = 298328, upload-time = "2025-10-14T15:03:54.281Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/ae/c6/cdeec138f6640bca687337291c90636ef7a43cf1a7147438f22d082a27fe/unsloth-2025.10.8.tar.gz", hash = "sha256:a35abeaf780c0bcfab7ee3f07a1fe3771f98b1a732ca937b71f3dffc7c929285", size = 4742041, upload-time = "2025-10-22T14:59:35.934Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/0d/d6/ba65db23d1da25e0943a73aa7fbc79ff6f98d84b93a4ebb986f6c3e7c3e8/unsloth-2025.10.3-py3-none-any.whl", hash = "sha256:e611f423cd62377aa78aaf259ee23763314c8dcdfd95535f747e245c6206b226", size = 337860, upload-time = "2025-10-14T15:03:48.931Z" },
+    { url = "https://files.pythonhosted.org/packages/c0/e2/65bebfc7e5881fd2531d3613b6a3585cb1262cad9bcd0d084597bc9b4159/unsloth-2025.10.8-py3-none-any.whl", hash = "sha256:0f86570c024d865493f05b4754fb2188c32da2132619c66f8b85b5c4e7a17428", size = 347307, upload-time = "2025-10-22T14:59:29.061Z" },
 ]
 
 [[package]]
 name = "unsloth-zoo"
-version = "2025.10.3"
+version = "2025.10.9"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "accelerate" },
@@ -7701,7 +7701,6 @@ dependencies = [
     { name = "filelock" },
     { name = "hf-transfer" },
     { name = "huggingface-hub" },
-    { name = "mistral-common" },
     { name = "msgspec" },
     { name = "numpy" },
     { name = "packaging" },
@@ -7722,9 +7721,9 @@ dependencies = [
     { name = "tyro" },
     { name = "wheel" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/82/37/566c5b89a323f1c3d8e3eadff57dc8b1a45a3faa9aa32debf82d555601e8/unsloth_zoo-2025.10.3.tar.gz", hash = "sha256:92c145c2734b5fda496af943fb31302d365d686e1a08325bd5b19f15cb5a108d", size = 242049, upload-time = "2025-10-14T14:56:08.579Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/7c/e9/c5e1ffd3a8cb5516387c72f86a5aa5b4f0536b6f9e45a899ce83406a3fcf/unsloth_zoo-2025.10.9.tar.gz", hash = "sha256:dbfb3094068a6dd488a8c786a0c681a5b6a26167e49ca990c1a04aebe1304cb2", size = 250615, upload-time = "2025-10-22T14:57:17.069Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/23/55/4e2e8646d9170ecce04977ed4e190a59607473ffb24a4108b885feac188a/unsloth_zoo-2025.10.3-py3-none-any.whl", hash = "sha256:8e90da9341c7c1a37253bd5f89490b56cb7d6aa1c65e63a981ce7b9b83d43464", size = 263520, upload-time = "2025-10-14T14:56:06.905Z" },
+    { url = "https://files.pythonhosted.org/packages/75/bd/984a148b147881d7833ae67bd1dc05b730b904833d5446f4b25d74bd64b5/unsloth_zoo-2025.10.9-py3-none-any.whl", hash = "sha256:06c1642af57cca5ec9b022d58972be1b41fc644f88efbd70b06c11471e75c54f", size = 269385, upload-time = "2025-10-22T14:57:15.389Z" },
 ]
 
 [[package]]