Qwen2.5-7B-Instruct-quantized.w8a8

Model Overview

• Model Architecture: Qwen2
  • Input: Text
  • Output: Text
• Model Optimizations:
  • Activation quantization: INT8
  • Weight quantization: INT8
• Intended Use Cases: Intended for commercial and research use multiple languages. Similarly to Qwen2.5-7B, this models is intended for assistant-like chat.
• Out-of-scope: Use in any manner that violates applicable laws or regulations (including trade compliance laws).
• Release Date: 10/09/2024
• Version: 1.0
• License(s): apache-2.0
• Model Developers: Neural Magic

Model Optimizations

This model was obtained by quantizing activations and weights of Qwen2.5-7B-Instruct to INT8 data type. This optimization reduces the number of bits used to represent weights and activations from 16 to 8, reducing GPU memory requirements (by approximately 50%) and increasing matrix-multiply compute throughput (by approximately 2x). Weight quantization also reduces disk size requirements by approximately 50%.

Only weights and activations of the linear operators within transformers blocks are quantized. Weights are quantized with a symmetric static per-channel scheme, whereas activations are quantized with a symmetric dynamic per-token scheme. A combination of the SmoothQuant and GPTQ algorithms is applied for quantization, as implemented in the llm-compressor library.

Deployment

This model can be deployed efficiently using the vLLM backend, as shown in the example below.

from vllm import LLM, SamplingParams
from transformers import AutoTokenizer

model_id = "RedHatAI/Qwen2.5-7B-Instruct-quantized.w8a8"
number_gpus = 1
max_model_len = 8192

sampling_params = SamplingParams(temperature=0.7, top_p=0.8, max_tokens=256)

tokenizer = AutoTokenizer.from_pretrained(model_id)

messages = [
    {"role": "user", "content": "Give me a short introduction to large language model."},
]

prompts = tokenizer.apply_chat_template(messages, tokenize=False)

llm = LLM(model=model_id, tensor_parallel_size=number_gpus, max_model_len=max_model_len)

outputs = llm.generate(prompts, sampling_params)

generated_text = outputs[0].outputs[0].text
print(generated_text)

vLLM aslo supports OpenAI-compatible serving. See the documentation for more details.

Creation

Creation details

This model was created with [llm-compressor](https://github.com/vllm-project/llm-compressor) by running the code snippet below.

from transformers import AutoModelForCausalLM, AutoTokenizer
from llmcompressor.modifiers.quantization import GPTQModifier
from llmcompressor.modifiers.smoothquant import SmoothQuantModifier  
from llmcompressor.transformers import oneshot
from datasets import load_dataset

# Load model
model_stub = "Qwen/Qwen2.5-7B-Instruct"
model_name = model_stub.split("/")[-1]

num_samples = 512
max_seq_len = 8192

tokenizer = AutoTokenizer.from_pretrained(model_stub)

model = AutoModelForCausalLM.from_pretrained(
    model_stub,
    device_map="auto",
    torch_dtype="auto",
)

def preprocess_fn(example):
    return {"text": tokenizer.apply_chat_template(example["messages"], add_generation_prompt=False, tokenize=False)}

ds = load_dataset("neuralmagic/LLM_compression_calibration", split="train")
ds = ds.map(preprocess_fn)

# Configure the quantization algorithm and scheme
recipe = [
    SmoothQuantModifier(
      smoothing_strength=0.8,
      mappings=[
          [["re:.*q_proj", "re:.*k_proj", "re:.*v_proj"], "re:.*input_layernorm"],
          [["re:.*gate_proj", "re:.*up_proj"], "re:.*post_attention_layernorm"],
          [["re:.*down_proj"], "re:.*up_proj"],
      ],
    ),
    GPTQModifier(
        ignore=["lm_head"],
        sequential_targets=["Qwen2DecoderLayer"],
        dampening_frac=0.01,
        targets="Linear",
        scheme="W8A8",
    ),
]

# Apply quantization
oneshot(
    model=model,
    dataset=ds, 
    recipe=recipe,
    max_seq_length=max_seq_len,
    num_calibration_samples=num_samples,
)

# Save to disk in compressed-tensors format
save_path = model_name + "-quantized.w8a8"
model.save_pretrained(save_path)
tokenizer.save_pretrained(save_path)
print(f"Model and tokenizer saved to: {save_path}")

Evaluation

The model was evaluated on the OpenLLM leaderboard tasks (version 1) with the lm-evaluation-harness (commit 387Bbd54bc621086e05aa1b030d8d4d5635b25e6) and the vLLM engine, using the following command:

lm_eval \
  --model vllm \
  --model_args pretrained="neuralmagic/Qwen2.5-7B-Instruct-quantized.w8a8",dtype=auto,gpu_memory_utilization=0.5,max_model_len=4096,enable_chunk_prefill=True,tensor_parallel_size=1 \
  --apply_chat_template \
  --fewshot_as_multiturn \
  --tasks openllm \
  --batch_size auto

Accuracy

Open LLM Leaderboard evaluation scores

Benchmark	Qwen2.5-7B-Instruct	Qwen2.5-7B-Instruct-quantized.w8a8 (this model)	Recovery
MMLU (5-shot)	74.24	73.87	99.5%
ARC Challenge (25-shot)	63.40	63.23	99.7%
GSM-8K (5-shot, strict-match)	80.36	80.74	100.5%
Hellaswag (10-shot)	81.52	81.06	99.4%
Winogrande (5-shot)	74.66	74.82	100.2%
TruthfulQA (0-shot, mc2)	64.76	64.58	99.7%
Average	73.16	73.05	99.4%