Zero-Shot Classification
Transformers
Safetensors
qwen3_5
feature-extraction
decision-model
classification
system-one
multimodal
vision
video
quantized
compressed-tensors
nvfp4
custom_code
8-bit precision
Instructions to use vllm-sr/d3-NVFP4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use vllm-sr/d3-NVFP4 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("zero-shot-classification", model="vllm-sr/d3-NVFP4", trust_remote_code=True)# pip install -U transformers accelerate # Load model directly from transformers import AutoProcessor, AutoModel processor = AutoProcessor.from_pretrained("vllm-sr/d3-NVFP4", trust_remote_code=True) model = AutoModel.from_pretrained("vllm-sr/d3-NVFP4", trust_remote_code=True, device_map="auto") - Notebooks
- Google Colab
- Kaggle
Commit History
3 days ago
-
Prepare for publishing: current d3_runtime.py, no local paths, drop dev eval files
Verified signature- d3_runtime.py: the version vllm-sr/d3 ships (its release verify skips README and other docs) - QUANT_INFO.json: source and calibration without local paths - eval_summary.json, eval_text_summary.json, eval_vision_summary.json removed (dev subsets the card does not cite) - d3q_engine.py usage line without local paths - README: tags, NVFP4 runtime now on main (#5225), wording - MODEL_MANIFEST.json updated for these files; identity unchanged95a92cc -
Card: serve with vllm-sr Engine mode (#5225, #5233 merged); merged projections
Verified signature49dc8b0
4 days ago
-
Add MODEL_MANIFEST.json for the vLLM Semantic Router model runtime
Verified signature446ce02
5 days ago
-
Card: latency served at --dtype auto (23.2 ms), exact-kernel columns kept
Verified signature0a86c1c
6 days ago
-
Card: tasksource test split through the model runtime (82.71%, 96.64% agreement)
Verified signature07fa5a0
7 days ago
-
Card: Decision Index, public vision suite and Perception Test through the model runtime with NVFP4 support (64.83, 73.60, 76.57%)
Verified signature90dfece -
Card: why NVFP4 needs graphs (kernel launches), shorter serving notes
Verified signature880ec9a -
Card: optimized NVFP4 layers next to the first ones (28.5 ms with graphs, 48.0 ms without), and what changed
Verified signature5cc266a -
Model card in the d3-FP8 card's layout: Decision Index per area with agreement, serving performance on RTX PRO 6000 with the NVFP4 runtime of #5145
Verified signaturec998473
9 days ago
-
Model card: drop tasksource eval600 (superseded by the full test split, which the runtime cannot run on NVFP4)
Verified signature05e5771 -
Model card: the vLLM Semantic Router runtime (#5066) loads FP8 and int4, not NVFP4; link d3-FP8 and d3-W4A16-G128-AWQ
Verified signaturede3214b -
Decision Index public suite: d3-NVFP4 complete (64.72, 96.96% agreement with BF16); text table with BF16 deltas
Verified signaturecce3779
10 days ago
-
Model card in the d3 card's layout; full-suite evaluation (in progress); add d3q_engine.py
Verified signature23d7073 -
NVFP4 (W4A4) quantization of vllm-sr/d3 with FP4 runtime kernels and eval results
Hsiung Li SyunVerified signature61b3f92 -
initial commit
Hsiung Li SyunVerified signaturecf91b69