imatrix-combine-only idea

[robbiemu]

llama-imatrix supports --in-file combination, but requires the user to also still process at least one chunk during that final step of combination. (otherwise, we get output like this:

/Users/macdev/Downloads/build/bin/llama-imatrix --model /Users/Shared/Public/huggingface/salamandra-2b-instruct/salamandra-2b-instruct_bf16.gguf \
  --rope-freq-base 10000.0 --top-p 0.95 --temp 0 --repeat-penalty 1.2 \
   --ctx-size 8192 --n-gpu-layers 25 --no-ppl --verbosity 3 \
  --in-file /Users/Shared/Public/huggingface/salamandra-2b-instruct/imatrix/oscar/imatrix_500.to_933.dat \
  --in-file /Users/Shared/Public/huggingface/salamandra-2b-instruct/imatrix/oscar/imatrix_500.from_933.dat \
  -o /Users/Shared/Public/huggingface/salamandra-2b-instruct/imatrix/oscar/imatrix_500.dat --chunk 0
build: 3906 (7eee341b) with Apple clang version 15.0.0 (clang-1500.3.9.4) for arm64-apple-darwin23.6.0
main : loading imatrix from '/Users/Shared/Public/huggingface/salamandra-2b-instruct/imatrix/oscar/imatrix_500.to_933.dat'
main : loading imatrix from '/Users/Shared/Public/huggingface/salamandra-2b-instruct/imatrix/oscar/imatrix_500.from_933.dat'
main : saving combined imatrix to '/Users/Shared/Public/huggingface/salamandra-2b-instruct/imatrix/oscar/imatrix_500.dat'

save_imatrix: stored collected data after 0 chunks in /Users/Shared/Public/huggingface/salamandra-2b-instruct/imatrix/oscar/imatrix_500.dat
llama_model_loader: loaded meta data with 31 key-value pairs and 219 tensors from /Users/Shared/Public/huggingface/salamandra-2b-instruct/salamandra-2b-instruct_bf16.gguf (version GGUF V3 (latest))
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
llama_model_loader: - kv   0:                       general.architecture str              = llama
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                         general.size_label str              = 2.3B
llama_model_loader: - kv   3:                            general.license str              = apache-2.0
llama_model_loader: - kv   4:                               general.tags arr[str,1]       = ["text-generation"]
llama_model_loader: - kv   5:                          general.languages arr[str,36]      = ["bg", "ca", "code", "cs", "cy", "da"...
llama_model_loader: - kv   6:                          llama.block_count u32              = 24
llama_model_loader: - kv   7:                       llama.context_length u32              = 8192
llama_model_loader: - kv   8:                     llama.embedding_length u32              = 2048
llama_model_loader: - kv   9:                  llama.feed_forward_length u32              = 5440
llama_model_loader: - kv  10:                 llama.attention.head_count u32              = 16
llama_model_loader: - kv  11:              llama.attention.head_count_kv u32              = 16
llama_model_loader: - kv  12:                       llama.rope.freq_base f32              = 10000.000000
llama_model_loader: - kv  13:     llama.attention.layer_norm_rms_epsilon f32              = 0.000010
llama_model_loader: - kv  14:                          general.file_type u32              = 32
llama_model_loader: - kv  15:                           llama.vocab_size u32              = 256000
llama_model_loader: - kv  16:                 llama.rope.dimension_count u32              = 128
llama_model_loader: - kv  17:            tokenizer.ggml.add_space_prefix bool             = true
llama_model_loader: - kv  18:                       tokenizer.ggml.model str              = llama
llama_model_loader: - kv  19:                         tokenizer.ggml.pre str              = default
llama_model_loader: - kv  20:                      tokenizer.ggml.tokens arr[str,256000]  = ["<unk>", "<s>", "</s>", "<pad>", "<|...
llama_model_loader: - kv  21:                      tokenizer.ggml.scores arr[f32,256000]  = [-1000.000000, -1000.000000, -1000.00...
llama_model_loader: - kv  22:                  tokenizer.ggml.token_type arr[i32,256000]  = [3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, ...
llama_model_loader: - kv  23:                tokenizer.ggml.bos_token_id u32              = 1
llama_model_loader: - kv  24:                tokenizer.ggml.eos_token_id u32              = 2
llama_model_loader: - kv  25:            tokenizer.ggml.unknown_token_id u32              = 0
llama_model_loader: - kv  26:            tokenizer.ggml.padding_token_id u32              = 0
llama_model_loader: - kv  27:               tokenizer.ggml.add_bos_token bool             = true
llama_model_loader: - kv  28:               tokenizer.ggml.add_eos_token bool             = false
llama_model_loader: - kv  29:                    tokenizer.chat_template str              = {%- if not date_string is defined %}{...
llama_model_loader: - kv  30:               general.quantization_version u32              = 2
llama_model_loader: - type  f32:   49 tensors
llama_model_loader: - type bf16:  170 tensors
llm_load_vocab: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
llm_load_vocab: special tokens cache size = 104
llm_load_vocab: token to piece cache size = 1.8842 MB
llm_load_print_meta: format           = GGUF V3 (latest)
llm_load_print_meta: arch             = llama
llm_load_print_meta: vocab type       = SPM
llm_load_print_meta: n_vocab          = 256000
llm_load_print_meta: n_merges         = 0
llm_load_print_meta: vocab_only       = 0
llm_load_print_meta: n_ctx_train      = 8192
llm_load_print_meta: n_embd           = 2048
llm_load_print_meta: n_layer          = 24
llm_load_print_meta: n_head           = 16
llm_load_print_meta: n_head_kv        = 16
llm_load_print_meta: n_rot            = 128
llm_load_print_meta: n_swa            = 0
llm_load_print_meta: n_embd_head_k    = 128
llm_load_print_meta: n_embd_head_v    = 128
llm_load_print_meta: n_gqa            = 1
llm_load_print_meta: n_embd_k_gqa     = 2048
llm_load_print_meta: n_embd_v_gqa     = 2048
llm_load_print_meta: f_norm_eps       = 0.0e+00
llm_load_print_meta: f_norm_rms_eps   = 1.0e-05
llm_load_print_meta: f_clamp_kqv      = 0.0e+00
llm_load_print_meta: f_max_alibi_bias = 0.0e+00
llm_load_print_meta: f_logit_scale    = 0.0e+00
llm_load_print_meta: n_ff             = 5440
llm_load_print_meta: n_expert         = 0
llm_load_print_meta: n_expert_used    = 0
llm_load_print_meta: causal attn      = 1
llm_load_print_meta: pooling type     = 0
llm_load_print_meta: rope type        = 0
llm_load_print_meta: rope scaling     = linear
llm_load_print_meta: freq_base_train  = 10000.0
llm_load_print_meta: freq_scale_train = 1
llm_load_print_meta: n_ctx_orig_yarn  = 8192
llm_load_print_meta: rope_finetuned   = unknown
llm_load_print_meta: ssm_d_conv       = 0
llm_load_print_meta: ssm_d_inner      = 0
llm_load_print_meta: ssm_d_state      = 0
llm_load_print_meta: ssm_dt_rank      = 0
llm_load_print_meta: ssm_dt_b_c_rms   = 0
llm_load_print_meta: model type       = ?B
llm_load_print_meta: model ftype      = BF16
llm_load_print_meta: model params     = 2.25 B
llm_load_print_meta: model size       = 4.20 GiB (16.00 BPW) 
llm_load_print_meta: general.name     = n/a
llm_load_print_meta: BOS token        = 1 '<s>'
llm_load_print_meta: EOS token        = 2 '</s>'
llm_load_print_meta: UNK token        = 0 '<unk>'
llm_load_print_meta: PAD token        = 0 '<unk>'
llm_load_print_meta: LF token         = 145 '<0x0A>'
llm_load_print_meta: EOT token        = 5 '<|im_end|>'
llm_load_print_meta: EOG token        = 2 '</s>'
llm_load_print_meta: EOG token        = 5 '<|im_end|>'
llm_load_print_meta: max token length = 72
llm_load_tensors: ggml ctx size =    0.20 MiB
ggml_backend_metal_log_allocated_size: allocated buffer, size =  4298.41 MiB, ( 4298.48 / 40960.00)
llm_load_tensors: offloading 24 repeating layers to GPU
llm_load_tensors: offloading non-repeating layers to GPU
llm_load_tensors: offloaded 25/25 layers to GPU
llm_load_tensors:      Metal buffer size =  4298.39 MiB
llm_load_tensors:        CPU buffer size =  1000.00 MiB
.......................................................
llama_new_context_with_model: n_ctx      = 8192
llama_new_context_with_model: n_batch    = 2048
llama_new_context_with_model: n_ubatch   = 512
llama_new_context_with_model: flash_attn = 0
llama_new_context_with_model: freq_base  = 10000.0
llama_new_context_with_model: freq_scale = 1
ggml_metal_init: allocating
ggml_metal_init: found device: Apple M3 Max
ggml_metal_init: picking default device: Apple M3 Max
ggml_metal_init: using embedded metal library
ggml_metal_init: GPU name:   Apple M3 Max
ggml_metal_init: GPU family: MTLGPUFamilyApple9  (1009)
ggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)
ggml_metal_init: GPU family: MTLGPUFamilyMetal3  (5001)
ggml_metal_init: simdgroup reduction support   = true
ggml_metal_init: simdgroup matrix mul. support = true
ggml_metal_init: hasUnifiedMemory              = true
ggml_metal_init: recommendedMaxWorkingSetSize  = 42949.67 MB
llama_kv_cache_init:      Metal KV buffer size =  1536.00 MiB
llama_new_context_with_model: KV self size  = 1536.00 MiB, K (f16):  768.00 MiB, V (f16):  768.00 MiB
llama_new_context_with_model:        CPU  output buffer size =     0.98 MiB
llama_new_context_with_model:      Metal compute buffer size =   288.00 MiB
llama_new_context_with_model:        CPU compute buffer size =   500.00 MiB
llama_new_context_with_model: graph nodes  = 774
llama_new_context_with_model: graph splits = 339

system_info: n_threads = 12 (n_threads_batch = 12) / 16 | AVX = 0 | AVX_VNNI = 0 | AVX2 = 0 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | AVX512_BF16 = 0 | FMA = 0 | NEON = 1 | SVE = 0 | ARM_FMA = 1 | F16C = 0 | FP16_VA = 1 | RISCV_VECT = 0 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 0 | SSSE3 = 0 | VSX = 0 | MATMUL_INT8 = 1 | LLAMAFILE = 1 | 
compute_imatrix: tokenizing the input ..
compute_imatrix: tokenization took 0.001 ms
compute_imatrix: you need at least 16384 tokens for a context of 8192 tokens
compute_imatrix: the data file you provided tokenizes to only 1 tokens

This is rather silly, so I've written a quick "please consider" pull request to indicate the desired behavior.

I believe this simple change is all that is needed because the prompt is pre-filled from arg.cpp, but not yet padded with control tokens until compute_imatrix calls the tokenize downstream

[robbiemu]

mi PR is obviously not effecting llama-serverso that fail probably isn't related. But I notice Im a couple of commits behind, let me know if you want me to catch up to latest @slaren (noticed you triggered the CI on this)