GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
axteam
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
axteam
Model Tree
Qwen/Qwen3.5-0.8B
Input Modalities
Output Modalities
Supported Functionality
pack-quantized.model.visual.*), the multi-token-prediction head (mtp.*), all RMSNorms, token embeddings, the LM head, the SSM non-linear parts (linear_attn.conv1d/A_log/dt_bias). Left unquantized to preserve quality (about 42% of all parameters in this build).chat_template / generation_config are identical to the original. config.json is the original plus a quantization_config block.This is post-training quantization (PTQ) — no retraining. AWQ uses a small calibration set (no gradient) to protect salient weight channels before rounding.
Every column was measured with the same harness and settings — lm-evaluation-harness on vLLM, greedy, few-shot completion (no chat template). Percentages are recovery vs. the BF16 original.
| Benchmark | BF16 (original) | FP8-dynamic | AWQ-8bit (this repo) | NVFP4A16 | NVFP4 | AWQ-4bit |
|---|---|---|---|---|---|---|
| MMLU (5-shot) | 50.56 | 50.50 (99.9%) | 50.31 (99.5%) | 48.39 (95.7%) | 46.36 (91.7%) | 48.55 (96.0%) |
| GSM8K (5-shot, strict) | 32.68 | 29.80 (91.2%) | 32.30 (98.8%) | 29.49 (90.3%) | 22.82 (69.8%) | 28.20 (86.3%) |
| ARC-Challenge (0-shot) | 37.54 | 36.52 (97.3%) | 37.46 (99.8%) | 35.07 (93.4%) | 33.11 (88.2%) | 35.15 (93.6%) |
| Winogrande (5-shot) | 56.27 | 55.49 (98.6%) | 56.67 (100.7%) | 55.80 (99.2%) | 55.64 (98.9%) | 55.88 (99.3%) |
| KMMLU (5-shot, direct) | 34.54 | 34.22 (99.1%) | 34.20 (99.0%) | 35.02 (101.4%) | 33.51 (97.0%) | 35.74 (103.5%) |
| HAE-RAE (0-shot) | 37.03 | 35.29 (95.3%) | 36.30 (98.0%) | 34.10 (92.1%) | 27.31 (73.8%) | 34.28 (92.6%) |
| Average | 41.44 | 40.30 (97.3%) | 41.21 (99.4%) | 39.65 (95.7%) | 36.46 (88.0%) | 39.63 (95.6%) |
Fidelity to the BF16 original on an internal mixed English/Korean set (100 prompts, BF16 greedy answers scored under each build — same harness for every row).
| Metric | This model | BF16 original |
|---|---|---|
| Perplexity (lower = closer) | 2.0200 | 2.0174 (+0.1%) |
| Top-1 next-token agreement | 98.1% | 98.7% |
These measure fidelity to the base model, not absolute task scores.
| Runtime | vLLM — the architecture is in vLLM ≥ 0.17.0; tested here with vLLM 0.32.0. |
| vLLM install | vLLM 0.32 itself needs Python 3.11–3.14, torch 2.13 (CUDA 13.0 build, so an NVIDIA driver with CUDA 13 support), transformers ≥5.16.1,<5.20 and compressed-tensors 0.17.0 (both pinned by vLLM; 0.17.0 reads this checkpoint, which was saved with 0.19.0 — do not upgrade it separately). |
| GPU | NVIDIA Turing (sm75) or newer — weight-only INT, no special tensor cores needed. V100 (sm70) is not supported. (Kernel choice per GPU from vLLM 0.32's selection code; only B200 was run here.) |
| Memory | Weights: 1.3 GB. Add KV cache (grows with context length × concurrent requests) and a few GB for activations / CUDA graphs; use --tensor-parallel-size N when the weights do not fit one GPU. |
| Transformers (without vLLM) | transformers>=5.2.0 + compressed-tensors>=0.15.0 (checked on CPU with Qwen3.5-0.8B FP8 / AWQ-4bit / NVFP4A16 / NVFP4). |
| trust_remote_code | not needed |
Start an OpenAI-compatible server:
bash
vllm serve axteam/Qwen3.5-0.8B-AWQ-8bit --max-model-len 32768Query it:
bash
curl http://localhost:8000/v1/chat/completions -H "Content-Type: application/json" -d '{ "model": "axteam/Qwen3.5-0.8B-AWQ-8bit", "messages": [{"role": "user", "content": "대한민국의 수도는 어디인가요?"}], "max_tokens": 512}'Offline inference in Python:
python
from vllm import LLM, SamplingParams llm = LLM(model="axteam/Qwen3.5-0.8B-AWQ-8bit", max_model_len=32768)params = SamplingParams(max_tokens=1024)out = llm.chat([{"role": "user", "content": "Explain model quantization in two sentences."}], params)print(out[0].outputs[0].text) # image input (URL or data: URI)messages = [{"role": "user", "content": [ {"type": "image_url", "image_url": {"url": "https://example.com/your-image.png"}}, {"type": "text", "text": "Describe this image."},]}]out = llm.chat(messages, params)print(out[0].outputs[0].text)Thinking mode is off by default in this model's chat template; to turn it on pass
chat_template_kwargs={"enable_thinking": True}(Pythonllm.chat(..., chat_template_kwargs=...), or"chat_template_kwargs": {"enable_thinking": true}in the request body). Text-only serving: add--limit-mm-per-prompt '{"image": 0, "video": 0}'to skip loading the vision encoder's memory budget.
The KV cache is kept in BF16 by default. KV-cache precision is a runtime choice, independent of the weight quantization above; for long contexts or large batches FP8 roughly halves KV-cache memory:
bash
vllm serve axteam/Qwen3.5-0.8B-AWQ-8bit --kv-cache-dtype fp8Quantized derivative of Qwen/Qwen3.5-0.8B (Apache-2.0). All original weights, tokenizer and chat template belong to the Qwen team.
Qwen/Qwen3.5-0.8B의 AWQ INT8 (W8A16) 양자화 빌드 · 제작 axteam
| 항목 | 내용 |
|---|---|
| 원본 모델 | Qwen/Qwen3.5-0.8B |
| 양자화 방식 | AWQ INT8 (W8A16) |
| 저장 형식 | compressed-tensors pack-quantized |
| 모델 구조 | 하이브리드 어텐션(선형 어텐션 + 풀 어텐션) · Dense · 멀티모달(비전) · MTP 헤드 |
| 용량 | 약 1.3 GB |
| 보정 데이터 | 샘플 256개 (한국어 위주 혼합 코퍼스) |
model.visual.*), MTP 헤드(mtp.*), 모든 RMSNorm, 토큰 임베딩, LM 헤드, SSM의 선형층 외 파라미터(linear_attn.conv1d·A_log·dt_bias) — 전체 파라미터의 약 42%config.json은 원본에 quantization_config 블록만 추가--tensor-parallel-size N으로 분산vllm serve axteam/Qwen3.5-0.8B-AWQ-8bit --max-model-len 32768 — OpenAI 호환 APILLM(model="axteam/Qwen3.5-0.8B-AWQ-8bit", max_model_len=32768).chat(...) — 전체 예시는 위 영어 본문 'Usage (vLLM)' 절 참고chat_template_kwargs={"enable_thinking": True} 전달(Python은 llm.chat() 인자, HTTP는 요청 본문의 "chat_template_kwargs": {"enable_thinking": true})image_url 항목 사용--limit-mm-per-prompt '{"image": 0, "video": 0}' — 비전 타워용 메모리 예약을 생략해 메모리 절약vllm serve에 --kv-cache-dtype fp8 추가 — 기본 BF16인 KV 캐시를 FP8로 저장해 KV 캐시 메모리를 약 절반으로 줄임 (긴 컨텍스트·동시 요청이 많을 때 유리, 가중치 양자화와 별개인 실행 옵션)curl http://localhost:8000/v1/chat/completions -H "Content-Type: application/json" -d '{ "model": "axteam/Qwen3.5-0.8B-AWQ-8bit", "messages": [{"role": "user", "content": "대한민국의 수도는 어디인가요?"}], "max_tokens": 512}'from vllm import LLM, SamplingParams llm = LLM(model="axteam/Qwen3.5-0.8B-AWQ-8bit", max_model_len=32768)params = SamplingParams(max_tokens=1024)out = llm.chat([{"role": "user", "content": "Explain model quantization in two sentences."}], params)print(out[0].outputs[0].text) # image input (URL or data: URI)messages = [{"role": "user", "content": [ {"type": "image_url", "image_url": {"url": "https://example.com/your-image.png"}}, {"type": "text", "text": "Describe this image."},]}]out = llm.chat(messages, params)print(out[0].outputs[0].text)