simaai/Qwen3-VL-2B-Instruct-GPTQ-Safetensors

Qwen3-VL-2B-Instruct GPTQ

This is a local post-training quantized checkpoint of Qwen/Qwen3-VL-2B-Instruct for later LLiMa compilation and deployment on Sima.ai hardware. It must not be uploaded as part of the current VLM batch.

Source revision: Not captured in the flattened local source directory. An immutable upstream revision must be recovered and recorded before any release or redistribution.

Source license metadata: Apache-2.0. This checkpoint remains subject to the upstream license, intended-use guidance, and limitations.

Quantization

The lm_head follows the default GPTQ INT4/G256 policy; no INT8 escalation was made. Calibration used lmms-lab/flickr30k test[:512], deterministic dataset order, 512 image-text samples, sequence length 2048, batch size 1, no concatenation, and no padding to maximum length. The exact target names and effective settings are saved in recipe.yaml.

Evaluation

Full MMStar used all 1,500 examples, VLMEvalKit commit 7055d3010c38ccb5dcae1bc9535ca19c7fe5d79f, deterministic generation, and exact_matching without an API judge.

Evaluation date: 2026-07-16. Slurm jobs: source 4651, quantized 4652. The aggregate evidence row is stored in vlmeval_results/mmstar_results.csv; raw predictions and status files remain under vlmeval_results/mmstar_full/.

Validation: MMStar inference passed (1,500/1,500).

Full matched WikiText-2 word perplexity (EleutherAI/wikitext_document_level, wikitext-2-raw-v1, no example limit; 2026-07-18): source 16.5131; this checkpoint 19.5099; absolute degradation +2.9968; relative degradation +18.15%. Raw JSON evidence: perplexity_results/vlm_full_wikitext/qwen3_vl_2b_{source,gptq_w4g256_visiongptq_w8}_v3.json.

Reproduction

This directory contains the exact quantize.py, recipe.yaml, and versions.txt used for the artifact:

python quantize.py \
  --model-path /path/to/source-model \
  --output-dir /path/to/new-quantized-model

Environment

Python: 3.13.2
torch: 2.11.0+cu128
CUDA: 12.8
transformers: 5.10.1
llm-compressor: 0.12.0
auto-round: 0.13.0
compressed-tensors: 0.17.1

Deployment

This is the pre-LLiMa quantized Hugging Face artifact. LLiMa compiler output must remain a separate deployment artifact unless its format and redistribution are approved. No Hugging Face upload is authorized for this batch.

Limitations

Quantization quality can vary by language, visual domain, prompt format, context length, and downstream runtime. The MMStar result does not replace application-specific validation. Audio components, when present, and explicitly excluded projectors remain in source precision as documented in recipe.yaml.