jan1k/Qwen3.6-35B-A3B-Uncensored-Genesis-Hermes-Final-NVFP4
Image-Text-to-Text ā¢ 35B ā¢ Updated ā¢ 521 ā¢ 1
Roy T
Aexand
AI & ML interests
None yet
Recent Activity
liked a model 6 days ago
jan1k/Qwen3.6-35B-A3B-Uncensored-Genesis-Hermes-Final-NVFP4 liked a model 6 days ago
jan1k/Qwen3.8-27B-Uncensored-Genesis-NVFP4 liked a model 6 days ago
Altworld/Hemmingway-1Organizations
None yet
liked 4
models 6 days ago
reacted to medmekk's post with š 6 days ago
Post
4324
š Introducing Halo 1.0
Today, we are open-sourcing Halo, the training framework we use to train every model at White Circle.
It comes with:
š§ Full post-training stack: SFT, DPO/KTO/SMPO, reward modeling, GRPO, distillation
š¤ Async multi-turn RL with vLLM/SGLang rollouts and sandboxed tool use
ā” ~2.8Ć TRL throughput on 8Ć B300 (EP+FSDPv2, FA4, fp8/fp4)
š¤ Dense HF models + 15 MoE families (Qwen, GLM, Mistral, DeepSeek-V4ā¦)
š ļø One halo command, prebuilt Docker images, and docs for humans and agents
š» https://github.com/whitecircle/halo
Try it and tell us what you're training
Today, we are open-sourcing Halo, the training framework we use to train every model at White Circle.
It comes with:
š§ Full post-training stack: SFT, DPO/KTO/SMPO, reward modeling, GRPO, distillation
š¤ Async multi-turn RL with vLLM/SGLang rollouts and sandboxed tool use
ā” ~2.8Ć TRL throughput on 8Ć B300 (EP+FSDPv2, FA4, fp8/fp4)
š¤ Dense HF models + 15 MoE families (Qwen, GLM, Mistral, DeepSeek-V4ā¦)
š ļø One halo command, prebuilt Docker images, and docs for humans and agents
š» https://github.com/whitecircle/halo
Try it and tell us what you're training
- 1 reply
liked 2
models 6 days ago
liked a
model 8 days ago
liked a
model 12 days ago
liked 2
models 14 days ago
liked a
model 18 days ago
liked a
model 19 days ago
liked 3
models 20 days ago
liked 2
models 21 days ago
liked 2
models 22 days ago
