llama : add AWQ for llama, llama2, mpt, and mistral models (#4593)

* update: awq support llama-7b model * update: change order * update: benchmark results for llama2-7b * update: mistral 7b v1 benchmark * update: support 4 models * fix: Readme * update: ready for PR * update: readme * fix: readme * update: change order import * black * format code * update: work for bot mpt and awqmpt * update: readme * Rename to llm_build_ffn_mpt_awq * Formatted other files * Fixed params count * fix: remove code * update: more detail for mpt * fix: readme * fix: readme * update: change folder architecture * fix: common.cpp * fix: readme * fix: remove ggml_repeat * update: cicd * update: cicd * uppdate: remove use_awq arg * update: readme * llama : adapt plamo to new ffn ggml-ci --------- Co-authored-by: Trần Đức Nam <v.namtd12@vinai.io> Co-authored-by: Le Hoang Anh <v.anhlh33@vinai.io> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
ggerganov · Dec 27, 2023 · f679349 · f679349
1 parent 879b690
commit f679349
Show file tree

Hide file tree

Showing 8 changed files with 443 additions and 5 deletions.
diff --git a/awq-py/README.md b/awq-py/README.md
@@ -0,0 +1,116 @@
+# AWQ: Activation-aware Weight Quantization for LLM - version apply to llamacpp
+[[Paper](https://arxiv.org/abs/2306.00978)][[Original Repo](https://github.com/mit-han-lab/llm-awq)][[Easy-to-use Repo](https://github.com/casper-hansen/AutoAWQ)]
+
+**Supported models:**
+
+- [X] LLaMA
+- [x] LLaMA 2
+- [X] MPT
+- [X] Mistral AI v0.1
+- [ ] Bloom
+- [ ] Mixtral MoE
+
+**TODO:**
+- [x] Update version work with both MPT and MPT-AWQ model
+- [ ] Add OPT model
+- [ ] Add Bloom model
+- [ ] Add Mixtral MoE
+- [ ] Support w3, w2
+
+
+## Contents
+
+- [Install](##Install)
+- [Convert](##Convert)
+- [Quantize](##Quantize)
+- [Test](##Test)
+- [Benchmark](##Benchmark)
+- [Results](##Results)
+
+## Install
+Install requirements
+```bash
+pip install -r requirements.txt
+```
+Get the pre-computed AWQ search results for multiple model families, including LLaMA, LLaMA2, MPT, OPT
+```bash
+git clone https://huggingface.co/datasets/mit-han-lab/awq-model-zoo awq_cache
+```
+
+## Convert
+Example for llama model
+```bash
+# For llama7b and llama2 models
+python convert.py models/llama-7b/ --awq-path awq_cache/llama-7b-w4-g128.pt --outfile models/llama_7b_fp16.gguf
+# For mistral and mpt models
+python convert-hf-to-gguf.py models/mpt-7b/ --awq-path awq_cache/llama-7b-w4-g128.pt --outfile models/mpt_7b_fp16.gguf
+```
+
+## Quantize
+```bash
+# We only benchmark and confirm the results on q4_0, q4_1, and q2_k types.
+./quantize models/llama_7b_fp16.gguf models/llama_7b_q4_0.gguf q4_0
+```
+
+## Test
+```bash
+# For all models.
+./build/bin/main -m models/llama_7b_q4_0.gguf -n 128 --prompt "Once upon a time"
+```
+
+## Benchmark
+The perplexity measurements in table above are done against the `wikitext2` test dataset (https://paperswithcode.com/dataset/wikitext-2), with context length of 512.
+```bash
+# For llama and llama2, and mistral models.
+./perplexity -m models/llama_7b_q4_0.gguf -f datasets/wikitext-2-raw/wiki.test.raw
+```
+
+## Results
+Results are run on OpenBLAS (CPU) and CuBLAS (GPU) for fair comparison
+We use three types of llamacpp quantization methods to work with our version, including q4_0, q4_1, and q2_k
+
+### Llama 7B (Build with OpenBLAS)
+
+| Model      | Measure      | F16    | Q4_0   | Q4_1   | Q2_K   |
+|-----------:|--------------|-------:|-------:|-------:|-------:|
+|Llama 7B    | perplexity   | 5.9066 | 6.1214 | 6.0643 | 6.5808 |
+|Llama 7B    | file size    |  12.9G  |   3.5G |   3.9G |   2.7G |
+|Llama 7B    | bits/weight  |   16.0 |    4.5 |    5.0 |    2.6 |
+|AWQ-LLama 7B| perplexity   | 5.9175 | 6.0252 | 5.9987 | 6.3692 |
+|AWQ-LLama 7B| file size    |  12.9G  |   3.5G |   3.9G |   2.7G |
+|AWQ-LLama 7B| bits/weight  |   16.0 |    4.5 |    5.0 |    2.6 |
+
+
+### Llama2 7B (Build with CuBLAS)
+
+| Model       | Measure      | F16    | Q4_0   | Q4_1   | Q2_K   |
+|------------:|--------------|-------:|-------:|-------:|-------:|
+|Llama2 7B    | perplexity   | 5.8664 | 6.0260 | 6.0656 | 6.4496 |
+|Llama2 7B    | file size    |  12.9G  |   3.5G |   3.9G |   2.7G |
+|Llama2 7B    | bits/weight  |   16.0 |    4.5 |    5.0 |    2.6 |
+|AWQ-LLama2 7B| perplexity   | 5.8801 | 6.0054 | 5.9849 | 6.3650 |
+|AWQ-LLama2 7B| file size    |  12.9G  |   3.5G |   3.9G |   2.7G |
+|AWQ-LLama2 7B| bits/weight  |   16.0 |    4.5 |    5.0 |    2.6 |
+
+
+### Mistral 7B v0.1 (Build with CuBLAS)
+
+| Model        | Measure      | F16    | Q4_0   | Q4_1   | Q2_K   |
+|-------------:|--------------|-------:|-------:|-------:|-------:|
+|Mistral 7B    | perplexity   | 5.6931 | 5.8202 | 5.8268 | 6.1645 |
+|Mistral 7B    | file size     |  14.5G |   4.1G |   4.5G |   3.1G |
+|Mistral 7B    | bits/weight  |   16.0 |    4.5 |    5.0 |    2.6 |
+|AWQ-Mistral 7B| perplexity   | 5.6934 | 5.8020 | 5.7691 | 6.0426 |
+|AWQ-Mistral 7B| file size     |  14.5G |   4.1G |   4.5G |   3.1G |
+|AWQ-Mistral 7B| bits/weight  |   16.0 |    4.5 |    5.0 |    2.6 |
+
+### MPT 7B (Build with OpenBLAS)
+
+| Model    | Measure      | F16    | Q4_0   | Q4_1   | Q2_K    |
+|---------:|--------------|-------:|-------:|-------:|--------:|
+|MPT 7B    | perplexity   | 8.4369 | 8.7956 | 8.6265 | 11.4913 |
+|MPT 7B    | file size    |  13.7G  |   3.9G |   4.3G |   2.8G  |
+|MPT 7B    | bits/weight  |   16.0 |    4.5 |    5.0 |    2.6  |
+|AWQ-MPT 7B| perplexity   | 8.4944 | 8.7053 |  8.6750 | 10.2873|
+|AWQ-MPT 7B| file size    |  13.7G  |   3.9G |   4.3G |   2.8G  |
+|AWQ-MPT 7B| bits/weight  |   16.0 |    4.5 |    5.0 |    2.6  |
diff --git a/awq-py/awq/apply_awq.py b/awq-py/awq/apply_awq.py
@@ -0,0 +1,254 @@
+"""
+Implements the AWQ for llama.cpp use cases.
+Original paper: https://arxiv.org/abs/2306.00978
+
+This code is based on versions of the AWQ implementation found in the following repositories:
+* https://github.com/mit-han-lab/llm-awq
+* https://github.com/casper-hansen/AutoAWQ
+"""
+
+import os
+import torch
+import torch.nn as nn
+
+from transformers import AutoModelForCausalLM, AutoConfig
+from transformers.models.bloom.modeling_bloom import BloomGelu
+from transformers.models.llama.modeling_llama import LlamaRMSNorm
+from transformers.activations import GELUActivation
+
+
+class ScaledActivation(nn.Module):
+    """
+    ScaledActivation module wraps an existing activation function and applies a
+    scale factor to its output.
+
+    Args:
+        module (nn.Module): The activation function to be scaled.
+        scales (torch.Tensor): A tensor of size (num_features,) containing the initial
+            scale factors for each feature.
+
+    Returns:
+        torch.Tensor: The scaled output of the activation function.
+    """
+
+    def __init__(self, module, scales):
+        super().__init__()
+        self.act = module
+        self.scales = nn.Parameter(scales.data)
+
+    def forward(self, x):
+        return self.act(x) / self.scales.view(1, 1, -1).to(x.device)
+
+
+def set_op_by_name(layer, name, new_module):
+    """
+    Set the new module for given module's name.
+
+    Args:
+        layer (nn.Module): The layer in which to replace the submodule.
+        name (str): The path to the submodule to be replaced, using dot notation
+            to access nested modules.
+        new_module (nn.Module): The new module to replace the existing one.
+    """
+    levels = name.split(".")
+    if len(levels) > 1:
+        mod_ = layer
+        for l_idx in range(len(levels) - 1):
+            if levels[l_idx].isdigit():
+                mod_ = mod_[int(levels[l_idx])]
+            else:
+                mod_ = getattr(mod_, levels[l_idx])
+        setattr(mod_, levels[-1], new_module)
+    else:
+        setattr(layer, name, new_module)
+
+
+def get_op_by_name(module, op_name):
+    """
+    Retrieves a submodule within a given layer based on its name.
+
+    Args:
+        module (nn.Module): The layer containing the submodule to find.
+        op_name (str): The name of the submodule.
+
+    Returns:
+        nn.Module: The requested submodule found within the given layer.
+
+    Raises:
+        ValueError: If the specified submodule cannot be found within the layer.
+    """
+    for name, m in module.named_modules():
+        if name == op_name:
+            return m
+    raise ValueError(f"Cannot find op {op_name} in module {module}")
+
+
+@torch.no_grad()
+def scale_ln_fcs(ln, fcs, scales):
+    """
+    Scales the weights of a LayerNorm and a list of fully-connected layers proportionally.
+
+    Args:
+        ln (nn.LayerNorm): The LayerNorm module to be scaled.
+        fcs (List[nn.Linear]): A list of fully-connected layers to be scaled.
+        scales (torch.Tensor): A 1D tensor of size (num_features,).
+    """
+
+    if not isinstance(fcs, list):
+        fcs = [fcs]
+
+    scales = scales.to(ln.weight.device)
+
+    ln.weight.div_(scales)
+    if hasattr(ln, "bias") and ln.bias is not None:
+        ln.bias.div_(scales)
+
+    for fc in fcs:
+        fc.weight.mul_(scales.view(1, -1))
+
+    for p in ln.parameters():
+        assert torch.isnan(p).sum() == 0
+    for fc in fcs:
+        for p in fc.parameters():
+            assert torch.isnan(p).sum() == 0
+
+
+@torch.no_grad()
+def scale_fc_fc(fc1, fc2, scales):
+    """
+    Scales the weights of two fully-connected layers in a specific pattern.
+
+    Args:
+        fc1 (nn.Linear): The first fully-connected layer to be scaled.
+        fc2 (nn.Linear): The second fully-connected layer to be scaled.
+        scales (torch.Tensor): A 1D tensor of size (num_features,).
+    """
+    assert isinstance(fc1, nn.Linear)
+    assert isinstance(fc2, nn.Linear)
+
+    scales = scales.to(fc1.weight.device)
+
+    fc1.weight[-scales.size(0):].div_(scales.view(-1, 1))
+    if fc1.bias is not None:
+        fc1.bias.div_(scales.view(-1))
+
+    fc2.weight.mul_(scales.view(1, -1))
+
+    for p in fc1.parameters():
+        assert torch.isnan(p).sum() == 0
+    for p in fc2.parameters():
+        assert torch.isnan(p).sum() == 0
+
+
+@torch.no_grad()
+def scale_gelu_fc(gelu, fc, scales):
+    """
+    Scales the weight of a GELU activation and a fully-connected layer proportionally.
+
+    Args:
+        gelu (Union[nn.GELU, BloomGelu, GELUActivation]): The GELU activation module to be scaled.
+        fc (nn.Linear): The fully-connected layer to be scaled.
+        scales (torch.Tensor): A 1D tensor of size (num_features,).
+
+    Raises:
+        TypeError: If the `gelu` module is not of type `nn.GELU`, `BloomGelu`, or `GELUActivation`.
+        TypeError: If the `fc` module is not of type `nn.Linear`.
+    """
+    assert isinstance(gelu, (nn.GELU, BloomGelu, GELUActivation))
+    assert isinstance(fc, nn.Linear)
+
+    fc.weight.mul_(scales.view(1, -1).to(fc.weight.device))
+
+    for p in fc.parameters():
+        assert torch.isnan(p).sum() == 0
+
+
+def apply_scale(module, scales_list, input_feat_dict=None):
+    """
+    Applies different scaling strategies to layers based on their type and hierarchy within a given module.
+
+    Args:
+        module (nn.Module): The module containing the layers to be scaled.
+        scales_list (List[Tuple[str, List[str], torch.Tensor]]): A list of tuples containing:
+            * prev_op_name (str): The name of the preceding operation or module,
+                relative to which the layers to be scaled are located.
+            * layer_names (List[str]): A list of names of the layers to be scaled, relative to the preceding operation.
+            * scales (torch.Tensor): A 1D tensor of size (num_features,) containing the scaling factors for each feature.
+        input_feat_dict (Optional[Dict[str, torch.Tensor]]): A dictionary mapping layer names to their corresponding
+            input features (optional).
+    """
+    for prev_op_name, layer_names, scales in scales_list:
+        prev_op = get_op_by_name(module, prev_op_name)
+        layers = [get_op_by_name(module, name) for name in layer_names]
+
+        prev_op.cuda()
+        for layer in layers:
+            layer.cuda()
+        scales.cuda()
+
+        if isinstance(prev_op, nn.Linear):
+            assert len(layers) == 1
+            scale_fc_fc(prev_op, layers[0], scales)
+        elif isinstance(prev_op, (nn.LayerNorm, LlamaRMSNorm)) or "rmsnorm" in str(prev_op.__class__).lower():
+            scale_ln_fcs(prev_op, layers, scales)
+        elif isinstance(prev_op, (nn.GELU, BloomGelu, GELUActivation)):
+            new_module = ScaledActivation(prev_op, scales)
+            set_op_by_name(module, prev_op_name, new_module)
+            scale_gelu_fc(prev_op, layers[0], scales)
+        else:
+            raise NotImplementedError(f"prev_op {type(prev_op)} not supported yet!")
+
+        # apply the scaling to input feat if given; prepare it for clipping
+        if input_feat_dict is not None:
+            for layer_name in layer_names:
+                inp = input_feat_dict[layer_name]
+                inp.div_(scales.view(1, -1).to(inp.device))
+
+        prev_op.cpu()
+        for layer in layers:
+            layer.cpu()
+        scales.cpu()
+
+
+@torch.no_grad()
+def apply_clip(module, clip_list):
+    """
+    Applies element-wise clipping to the weight of a specific layer within a given module.
+
+    Args:
+        module (nn.Module): The module containing the layer to be clipped.
+        clip_list (List[Tuple[str, torch.Tensor]]): A list of tuples containing:
+            * name (str): The name of the layer to be clipped, relative to the root of the module.
+            * max_val (torch.Tensor): A 1D or 2D tensor defining the upper bound for each element of the layer's weight.
+    """
+    for name, max_val in clip_list:
+        layer = get_op_by_name(module, name)
+        layer.cuda()
+        max_val = max_val.to(layer.weight.device)
+        org_shape = layer.weight.shape
+        layer.weight.data = layer.weight.data.reshape(*max_val.shape[:2], -1)
+        layer.weight.data = torch.clamp(layer.weight.data, -max_val, max_val)
+        layer.weight.data = layer.weight.data.reshape(org_shape)
+        layer.cpu()
+
+
+def add_scale_weights(model_path, scale_path, tmp_path):
+    """
+    Adds pre-computed Activation Weight Quantization (AWQ) results to a model,
+    including scaling factors and clipping bounds.
+
+    Args:
+        model_path (str): Path to the pre-trained model to be equipped with AWQ.
+        scale_path (str): Path to the AWQ scale factors (.pt file).
+        tmp_path (str): Path to the temporary directory where the equipped model will be saved.
+    """
+    config = AutoConfig.from_pretrained(model_path, trust_remote_code=True)
+    model = AutoModelForCausalLM.from_pretrained(
+        model_path, config=config, trust_remote_code=True
+    )
+    model.eval()
+    awq_results = torch.load(str(scale_path), map_location="cpu")
+    apply_scale(model, awq_results["scale"])
+    apply_clip(model, awq_results["clip"])
+    model.save_pretrained(str(tmp_path))
+    os.system(f"cp {str(model_path)}/tokenizer* {str(tmp_path)}")
diff --git a/awq-py/requirements.txt b/awq-py/requirements.txt
@@ -0,0 +1,2 @@
+torch>=2.0.0
+transformers>=4.32.0