DeepSeek-V4-Flash-0731-FP8-DSpark
FP8 re-packaging of deepseek-ai/DeepSeek-V4-Flash-0731 for Hopper (H200/H100).
Routed MoE experts converted MXFP4/FP4 -> block FP8 via lossless
cast_e2m1fn_to_e4m3fn (tmp/scripts/convert_hf_mxfp4_experts_to_fp8.py).
expert_dtype removed from config.json; other fields (including dspark_* if
present) kept.
On Hopper set SGLANG_DSV4_FP4_EXPERTS=0 when serving.
- Downloads last month
- 272,951
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for AtlasCloud/DeepSeek-V4-Flash-0731-FP8-DSpark
Base model
deepseek-ai/DeepSeek-V4-Flash-0731