Hy3-NVFP4

Hy3 NVFP4 by Tiyuvta

Weight-only NVFP4 safetensors quantization of Tencent Hy3, generated directly from the official BF16 checkpoint with NVIDIA Model Optimizer 0.46.0.

Minted by Tiyuvta (inference.tiyuvta.ai) and qualified on the tiyuvta serving engine.

Artifact availability is not hosted-model availability. Tiyuvta does not currently publish a hosted Hy3 model id. This repository is an open model artifact, not a customer endpoint.

Provenance

The tokenizer, chat template, generation config, and license come from the pinned source. The official Tencent FP8 checkpoint is a separately pinned control, not the source for this mint.

Precision split

There is no BF16-MTP variant or fallback in this release. The complete artifact contains 46,080 NVFP4 expert weights, 139,298 stored tensors normalized to 47,138 logical tensors, and an indexed tensor payload of 180,826,481,152 bytes.

Integrity locks

The repository includes SHA256SUMS for uploaded-byte verification. The minted-manifest hash above binds the quantizer output before the Hugging Face model card replaces the upstream README.md; the upstream README is retained as README_UPSTREAM.md.

Validation

The exact artifact was qualified on the serving engine on four NVIDIA RTX PRO 6000 Blackwell Server Edition cards. The serving engine's model-family-independent automatic loader compiled the canonical ModelPlan, bound the exact tensor census, and selected whole-expert parallelism for the four-card artifact. No HY3 layer list or architecture-specific parallel recipe was supplied.

• Same-artifact ModelOpt oracle (vLLM pipeline placement): matching argmax, 20/20 top-logit overlap, cosine similarity 0.9995285974311506, RMSE 0.05849238475772628, mean absolute error 0.046138540558894096, and maximum absolute error 0.30402064323425293.
• Automatic capacity planning selected PP-2 at the legal layer-41 cut, then EP-3 and EP-4 as the expert root fit with reserve. On four cards it estimated 60.05 GB on the root and 40.26 GB on each peer before the 6 GiB runtime reserve.
• Greedy MTP K=1–8: target-identical at every K, with nonzero accepted drafts at every K. The sampled server also engaged MTP explicitly (9/22 accepted in the qualification request).
• Vendor-default sampling was tested with no sampling fields; the serving engine resolved Tencent's temperature 0.9 and top-p 1.0 defaults.
• Tool-call parsing, tool-result turns, reasoning_effort high/none, concurrent requests, cache reuse, admission, client-disconnect rollback, peer integrity, and clean rollback to the plain server all passed.

MTP is qualified but remains an explicit serving choice. An experimental wave schedule also remains off for this artifact: it improved throughput internally, but failed the required serial-vs-wave logit-identity gate for the mixed BF16/NVFP4 numeric program. No internal aggregate throughput result is promoted into a public performance claim.

Tuning on the serving engine is not finished. This artifact is qualified, not yet tuned; generic 2–4 card, TP+EP, routing, and MTP work remains.

Qualified runtime

The artifact was qualified on the serving engine with all four cards visible, automatic parallel placement, speculative decoding off and batch size 1. Qualified runtime binaries:

The published sampling defaults follow Tencent's configuration: temperature 0.9 and top-p 1.0. Clients can override them per request.

Files

• 99 indexed safetensors shards;
• config.json, hf_quant_config.json, model.safetensors.index.json;
• tokenizer, tokenizer config, chat template, and generation config;
• Apache-2.0 license and retained upstream READMEs;
• SHA256SUMS covering the uploaded files (excluding itself).

License

Hy3 is licensed under Apache-2.0. This artifact carries the pinned upstream license.

Credits

• Tencent Hy Team for Hy3.
• NVIDIA for ModelOpt and the NVFP4 format.
• Tiyuvta for the serving engine (loader and tensor contract), artifact generation, and RTX Blackwell qualification.