Configuration Parsing Warning:In config.json: "quantization_config.modules_to_not_convert" must be an array
Qwen3-8B AWQ 4-bit
4-bit AWQ (Activation-aware Weight Quantization) of Qwen/Qwen3-8B.
Quantization details
| Parameter | Value |
|---|---|
| Method | AWQ (GEMM kernel) |
| Bits | 4 |
| Group size | 128 |
| Calibration | WikiText-2 (512 samples) |
Benchmark results (200-sample subset)
| Benchmark | BF16 Baseline | AWQ 4-bit | Drop |
|---|---|---|---|
| GSM8K accuracy | 19.0% | pending | — |
| BoolQ accuracy | 87.5% | pending | — |
| MS MARCO ROUGE-L | 0.0616 | pending | — |
AWQ benchmark evaluation in progress. Results will be updated once complete.
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained(
"shantipriya/Qwen3-8B-AWQ-4bit",
device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("shantipriya/Qwen3-8B-AWQ-4bit")
inputs = tokenizer("Hello, how are you?", return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
Citation
- Downloads last month
- 20