Merge pull request #7 from neuralmagic/ignore-layers

Add `ignore_patterns` arg for ignoring layers

Author: mgoin

auto_fp8/__init__.py

@@ -1,5 +1,5 @@
-from .modeling import AutoFP8ForCausalLM
 from .config import BaseQuantizeConfig
+from .modeling import AutoFP8ForCausalLM
 
 __all__ = [
     "AutoFP8ForCausalLM",

auto_fp8/config.py

@@ -1,5 +1,13 @@
+from typing import List
+
+
 class BaseQuantizeConfig:
-    def __init__(self, quant_method="fp8", activation_scheme="static"):
+    def __init__(
+        self,
+        quant_method: str = "fp8",
+        activation_scheme: str = "static",
+        ignore_patterns: List[str] = [],
+    ):
         if quant_method != "fp8":
             raise ValueError("Only FP8 quantization is supported.")
         if activation_scheme not in ["static", "dynamic"]:
@@ -8,3 +16,5 @@ def __init__(self, quant_method="fp8", activation_scheme="static"):
             )
         self.quant_method = quant_method
         self.activation_scheme = activation_scheme
+        self.ignore_patterns = ignore_patterns
+        self.ignored_layers = []

auto_fp8/modeling.py

@@ -1,24 +1,34 @@
+import re
+from typing import List
+
 import torch
-from transformers import AutoModelForCausalLM, PreTrainedModel
+from transformers import AutoModelForCausalLM
+
+from auto_fp8.config import BaseQuantizeConfig
 from auto_fp8.quantize import (
-    quantize_weights,
     quantize_activations,
+    quantize_weights,
     save_quantized_model,
 )
-from auto_fp8.config import BaseQuantizeConfig
 
 
 class AutoFP8ForCausalLM:
     def __init__(
         self,
-        model: PreTrainedModel,
+        model: AutoModelForCausalLM,
         quantize_config: BaseQuantizeConfig,
     ):
         self.model = model
         self.model_type = self.model.config.model_type
-        self.quantize_config = quantize_config
         self.config = self.model.config
 
+        # Gather the Linear module names that we want to ignore
+        quantize_config.ignored_layers = get_layers_to_ignore(
+            self.model, quantize_config.ignore_patterns
+        )
+
+        self.quantize_config = quantize_config
+
     @classmethod
     def from_pretrained(
         cls,
@@ -94,16 +104,47 @@ def _prepare_calibration_data(calibration_tokens):
             return calibration_tokens
 
         # Always quantize the weights as they do not require calibration data
-        quantize_weights(self.model)
+        quantize_weights(self.model, self.quantize_config)
 
         if self.quantize_config.activation_scheme == "static":
             quantize_activations(
-                self.model, _prepare_calibration_data(calibration_tokens)
+                self.model,
+                self.quantize_config,
+                _prepare_calibration_data(calibration_tokens),
             )
 
+            # import copy
+            # for layer in self.model.model.layers:
+            #     layer.self_attn.kv_scale = copy.deepcopy(layer.self_attn.k_proj.act_scale)
+
     def save_quantized(self, save_dir):
         save_quantized_model(
             self.model,
-            activation_scheme=self.quantize_config.activation_scheme,
+            quant_config=self.quantize_config,
             save_dir=save_dir,
         )
+
+
+def get_layers_to_ignore(model, ignore_patterns) -> List[str]:
+    ignored_layers = set()
+
+    # TODO: don't always ignore lm_head
+    ignore_patterns.append("re:.*lm_head")
+
+    for name, linear in model.named_modules():
+        if not isinstance(linear, torch.nn.Linear):
+            continue
+
+        for ignore_pattern in ignore_patterns:
+            regex_prefix = "re:"
+            if ignore_pattern.startswith(regex_prefix):
+                # check if name matches regex and add to set if true
+                regex_pattern = ignore_pattern[len(regex_prefix) :]
+                if re.search(regex_pattern, name):
+                    ignored_layers.add(name)
+            else:
+                # else, exact match
+                if ignore_pattern == name:
+                    ignored_layers.add(name)
+
+    return list(ignored_layers)

auto_fp8/quantize.py

@@ -1,10 +1,13 @@
 import gc
 import re
-from typing import Tuple
+from typing import List, Tuple
+
 import torch
-import transformers
 import tqdm
-from transformers import AutoTokenizer
+import transformers
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+from .config import BaseQuantizeConfig
 
 
 # HACK: Override the dtype_byte_size function in transformers to support float8 types
@@ -39,8 +42,8 @@ def per_tensor_quantize(tensor: torch.Tensor) -> Tuple[torch.Tensor, float]:
     if tensor.numel() == 0:
         # Deal with empty tensors (triggered by empty MoE experts)
         min_val, max_val = (
-            torch.tensor(0.0, dtype=tensor.dtype),
-            torch.tensor(1.0, dtype=tensor.dtype),
+            torch.tensor(-16.0, dtype=tensor.dtype),
+            torch.tensor(16.0, dtype=tensor.dtype),
         )
     else:
         min_val, max_val = tensor.aminmax()
@@ -80,7 +83,9 @@ def fp8_gemm(A, A_scale, B, B_scale, bias, out_dtype):
 
 
 class FP8StaticLinearQuantizer(torch.nn.Module):
-    def __init__(self, qweight, weight_scale, bias):
+    def __init__(
+        self, qweight: torch.Tensor, weight_scale: torch.Tensor, bias: torch.Tensor
+    ):
         super().__init__()
         self.weight = torch.nn.Parameter(qweight, requires_grad=False)
         self.weight_scale = torch.nn.Parameter(weight_scale, requires_grad=False)
@@ -105,7 +110,13 @@ def forward(self, x):
 
 
 class FP8StaticLinear(torch.nn.Module):
-    def __init__(self, qweight, weight_scale, bias, act_scale=0.0):
+    def __init__(
+        self,
+        qweight: torch.Tensor,
+        weight_scale: torch.Tensor,
+        bias: torch.Tensor,
+        act_scale: float = 1.0,
+    ):
         super().__init__()
         self.weight = torch.nn.Parameter(qweight, requires_grad=False)
         self.weight_scale = torch.nn.Parameter(weight_scale, requires_grad=False)
@@ -133,7 +144,7 @@ def forward(self, x):
 
 
 class FP8DynamicLinear(torch.nn.Module):
-    def __init__(self, qweight, scale, bias):
+    def __init__(self, qweight: torch.Tensor, scale: torch.Tensor, bias: torch.Tensor):
         super().__init__()
         self.weight = torch.nn.Parameter(qweight, requires_grad=False)
         self.weight_scale = torch.nn.Parameter(scale, requires_grad=False)
@@ -152,21 +163,28 @@ def forward(self, x):
         return output
 
 
-def replace_module(model, name, new_module):
+def replace_module(model: AutoModelForCausalLM, name: str, new_module: torch.nn.Module):
     if "." in name:
         parent_name = name.rsplit(".", 1)[0]
         child_name = name[len(parent_name) + 1 :]
-        parent = model.model.get_submodule(parent_name)
+        parent = model.get_submodule(parent_name)
     else:
         parent_name = ""
-        parent = model.model
+        parent = model
         child_name = name
     setattr(parent, child_name, new_module)
 
 
-def quantize_weights(model):
-    for name, linear in model.model.named_modules():
-        if not isinstance(linear, torch.nn.Linear):
+def quantize_weights(
+    model: AutoModelForCausalLM,
+    quantize_config: BaseQuantizeConfig,
+    ignored_layers: List[str] = [],
+):
+    for name, linear in model.named_modules():
+        if (
+            not isinstance(linear, torch.nn.Linear)
+            or name in quantize_config.ignored_layers
+        ):
             continue
         quant_weight, quant_scale = per_tensor_quantize(linear.weight)
         quant_linear = FP8DynamicLinear(quant_weight, quant_scale, linear.bias)
@@ -175,9 +193,17 @@ def quantize_weights(model):
     cleanup_memory()
 
 
-def quantize_activations(model, calibration_tokens):
-    for name, dynamic_quant_linear in model.model.named_modules():
-        if not isinstance(dynamic_quant_linear, FP8DynamicLinear):
+def quantize_activations(
+    model: AutoModelForCausalLM,
+    quantize_config: BaseQuantizeConfig,
+    calibration_tokens,
+    ignored_layers: List[str] = [],
+):
+    for name, dynamic_quant_linear in model.named_modules():
+        if (
+            not isinstance(dynamic_quant_linear, FP8DynamicLinear)
+            or name in quantize_config.ignored_layers
+        ):
             continue
         quantizer = FP8StaticLinearQuantizer(
             dynamic_quant_linear.weight,
@@ -196,8 +222,11 @@ def quantize_activations(model, calibration_tokens):
             pbar.update(1)
 
     # Replace dynamic quantizer with StaticLinear for export
-    for name, quantizer in model.model.named_modules():
-        if not isinstance(quantizer, FP8StaticLinearQuantizer):
+    for name, quantizer in model.named_modules():
+        if (
+            not isinstance(quantizer, FP8StaticLinearQuantizer)
+            or name in quantize_config.ignored_layers
+        ):
             continue
         static_proj = FP8StaticLinear(
             quantizer.weight,
@@ -210,13 +239,19 @@ def quantize_activations(model, calibration_tokens):
     cleanup_memory()
 
 
-def save_quantized_model(model, activation_scheme, save_dir):
+def save_quantized_model(
+    model: AutoModelForCausalLM,
+    quant_config: BaseQuantizeConfig,
+    save_dir: str,
+    ignored_layers: List[str] = [],
+):
     print(model)
     print(f"Saving the model to {save_dir}")
     static_q_dict = {
         "quantization_config": {
             "quant_method": "fp8",
-            "activation_scheme": activation_scheme,
+            "activation_scheme": quant_config.activation_scheme,
+            "ignored_layers": quant_config.ignored_layers,
         }
     }
     model.config.update(static_q_dict)

example.py

@@ -1,4 +1,5 @@
 from transformers import AutoTokenizer
+
 from auto_fp8 import AutoFP8ForCausalLM, BaseQuantizeConfig
 
 pretrained_model_dir = "meta-llama/Meta-Llama-3-8B-Instruct"
@@ -9,8 +10,10 @@
 examples = tokenizer(examples, return_tensors="pt").to("cuda")
 
 quantize_config = BaseQuantizeConfig(
-    quant_method="fp8", activation_scheme="dynamic"
-)  # or "static"
+    quant_method="fp8",
+    activation_scheme="dynamic",  # or "static"
+    ignore_patterns=["re:.*lm_head"],
+)
 
 model = AutoFP8ForCausalLM.from_pretrained(
     pretrained_model_dir, quantize_config=quantize_config

example_dataset.py

@@ -0,0 +1,22 @@
+from datasets import load_dataset
+from transformers import AutoTokenizer
+
+from auto_fp8 import AutoFP8ForCausalLM, BaseQuantizeConfig
+
+pretrained_model_dir = "meta-llama/Meta-Llama-3-8B-Instruct"
+quantized_model_dir = "Meta-Llama-3-8B-Instruct-FP8"
+
+tokenizer = AutoTokenizer.from_pretrained(pretrained_model_dir, use_fast=True)
+tokenizer.pad_token = tokenizer.eos_token
+
+ds = load_dataset("mgoin/ultrachat_2k", split="train_sft").select(512)
+examples = [tokenizer.apply_chat_template(batch["messages"], tokenize=False) for batch in ds]
+examples = tokenizer(examples, padding=True, truncation=True, return_tensors="pt").to("cuda")
+
+quantize_config = BaseQuantizeConfig(quant_method="fp8", activation_scheme="static")
+
+model = AutoFP8ForCausalLM.from_pretrained(
+    pretrained_model_dir, quantize_config=quantize_config
+)
+model.quantize(examples)
+model.save_quantized(quantized_model_dir)

examples/example_mixtral.py

@@ -0,0 +1,26 @@
+from datasets import load_dataset
+from transformers import AutoTokenizer
+
+from auto_fp8 import AutoFP8ForCausalLM, BaseQuantizeConfig
+
+pretrained_model_dir = "mistralai/Mixtral-8x7B-Instruct-v0.1"
+quantized_model_dir = "Mixtral-8x7B-Instruct-v0.1-FP8"
+
+tokenizer = AutoTokenizer.from_pretrained(pretrained_model_dir, use_fast=True)
+tokenizer.pad_token = tokenizer.eos_token
+
+ds = load_dataset("mgoin/ultrachat_2k", split="train_sft").select(range(10))
+examples = [tokenizer.apply_chat_template(batch["messages"], tokenize=False) for batch in ds]
+examples = tokenizer(examples, padding=True, truncation=True, return_tensors="pt").to("cuda")
+
+quantize_config = BaseQuantizeConfig(
+    quant_method="fp8",
+    activation_scheme="static",
+    ignore_patterns=["re:.*lm_head", "re:.*gate"],
+)
+
+model = AutoFP8ForCausalLM.from_pretrained(
+    pretrained_model_dir, quantize_config=quantize_config
+)
+model.quantize(examples)
+model.save_quantized(quantized_model_dir)

examples/quantize.py

@@ -4,8 +4,8 @@
 from typing import Tuple
 
 import torch
-import transformers
 import tqdm
+import transformers
 from datasets import load_dataset
 from transformers import AutoModelForCausalLM, AutoTokenizer
 

setup.py

@@ -1,4 +1,4 @@
-from setuptools import setup, find_packages
+from setuptools import find_packages, setup
 
 setup(
     name="auto_fp8",

tests/test_auto_fp8.py

@@ -1,7 +1,9 @@
 import os
+import shutil
+
 from transformers import AutoTokenizer
+
 from auto_fp8 import AutoFP8ForCausalLM, BaseQuantizeConfig
-import shutil
 
 
 def test_quantization():