NLA Models
Collection
Natural Language Autoencoders across Llama, Gemma, and Qwen. Code: https://github.com/kitft/natural_language_autoencoders/ β’ 8 items β’ Updated β’ 18
Dokyoon
leeloolee
AI & ML interests
ai
Recent Activity
upvoted a collection 10 days ago
NLA Models liked a model 16 days ago
yandex/AliceAI-Foundation-80B-A3B-BaseOrganizations
upvoted a collection 10 days ago
reacted to medmekk's post with π 16 days ago
Post
4363
π 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:
π§ Full post-training stack: SFT, DPO/KTO/SMPO, reward modeling, GRPO, distillation
π€ Async multi-turn RL with vLLM/SGLang rollouts and sandboxed tool use
β‘ ~2.8Γ TRL throughput on 8Γ B300 (EP+FSDPv2, FA4, fp8/fp4)
π€ Dense HF models + 15 MoE families (Qwen, GLM, Mistral, DeepSeek-V4β¦)
π οΈ One halo command, prebuilt Docker images, and docs for humans and agents
π» 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:
π§ Full post-training stack: SFT, DPO/KTO/SMPO, reward modeling, GRPO, distillation
π€ Async multi-turn RL with vLLM/SGLang rollouts and sandboxed tool use
β‘ ~2.8Γ TRL throughput on 8Γ B300 (EP+FSDPv2, FA4, fp8/fp4)
π€ Dense HF models + 15 MoE families (Qwen, GLM, Mistral, DeepSeek-V4β¦)
π οΈ One halo command, prebuilt Docker images, and docs for humans and agents
π» https://github.com/whitecircle/halo
Try it and tell us what you're training
- 1 reply
liked a
model 16 days ago
liked a
model 18 days ago
liked a
model 19 days ago
liked a
model 29 days ago
liked a
model about 1 month ago
liked a
model 3 months ago
liked 2
models 4 months ago
upvoted a paper 4 months ago
liked a
model 4 months ago
liked a
dataset 4 months ago
liked a
Space 4 months ago
liked 2
datasets 4 months ago
reacted to imnotkitty's post with π₯ 6 months ago
Post
4062
tencent/Hy3-preview is out: an open-weights MoE reasoning model.
β 295B total / 21B active / 256K context
β Fused fast-and-slow thinking in a single model
β First model trained on Hunyuan's rebuilt pretraining + RL infra (Feb β Apr)
Benchmarks:
π SWE-Bench Verified, Terminal-Bench 2.0, BrowseComp, WideSearch β competitive results, particularly strong on agentic tool use
π Top score on Tsinghua's 2026 Spring math PhD qualifying exam
π Strong context-learning and instruction-following on Tencent's CL-bench / CL-bench-Life
More details can be found in my article: https://huggingface.co/blog/imnotkitty/hy3-preview
β 295B total / 21B active / 256K context
β Fused fast-and-slow thinking in a single model
β First model trained on Hunyuan's rebuilt pretraining + RL infra (Feb β Apr)
Benchmarks:
π SWE-Bench Verified, Terminal-Bench 2.0, BrowseComp, WideSearch β competitive results, particularly strong on agentic tool use
π Top score on Tsinghua's 2026 Spring math PhD qualifying exam
π Strong context-learning and instruction-following on Tencent's CL-bench / CL-bench-Life
More details can be found in my article: https://huggingface.co/blog/imnotkitty/hy3-preview
- 2 replies
upvoted a paper 6 months ago
liked a
model 6 months ago
reacted to anakin87's post with β€οΈ 6 months ago
Post
3318
π£ I just published a free course on Reinforcement Learning Environments for Language Models!
π COURSE: https://github.com/anakin87/llm-rl-environments-lil-course
Over the past year, we've seen a shift in LLM Post-Training.
Previously, Supervised Fine-Tuning was the most important part: making models imitate curated Question-Answer pairs.
Now we also have Reinforcement Learning with Verifiable Rewards. With techniques like GRPO, models can learn through trial and error in dynamic environments. They can climb to new heights without relying on expensively prepared data.
But what actually are these environments in practiceβ And how do you build them effectivelyβ
Fascinated by these concepts, I spent time exploring this space through experiments, post-training Small Language Models.
I've packaged everything I learned into this short course.
What you'll learn
πΉ Agents, Environments, and LLMs: how to map Reinforcement Learning concepts to the LLM domain
πΉ How to use Verifiers (open-source library by Prime Intellect) to build RL environments as software artifacts
πΉ Common patterns: How to build single-turn, multi-turn, and tool-use environments
πΉ Hands-on: turn a small language model (LFM2-2.6B by LiquidAI) into a Tic Tac Toe master
πΈ Build the game Environment
πΈ Use it to generate synthetic data for SFT warm-up
πΈ Group-based Reinforcement Learning
If you're interested in building "little worlds" where LLMs can learn, this course is for you.
---
π€πΉοΈ Play against the trained model: anakin87/LFM2-2.6B-mr-tictactoe
π HF collection (datasets + models): https://huggingface.co/collections/anakin87/lfm2-26b-mr-tic-tac-toe
π COURSE: https://github.com/anakin87/llm-rl-environments-lil-course
Over the past year, we've seen a shift in LLM Post-Training.
Previously, Supervised Fine-Tuning was the most important part: making models imitate curated Question-Answer pairs.
Now we also have Reinforcement Learning with Verifiable Rewards. With techniques like GRPO, models can learn through trial and error in dynamic environments. They can climb to new heights without relying on expensively prepared data.
But what actually are these environments in practiceβ And how do you build them effectivelyβ
Fascinated by these concepts, I spent time exploring this space through experiments, post-training Small Language Models.
I've packaged everything I learned into this short course.
What you'll learn
πΉ Agents, Environments, and LLMs: how to map Reinforcement Learning concepts to the LLM domain
πΉ How to use Verifiers (open-source library by Prime Intellect) to build RL environments as software artifacts
πΉ Common patterns: How to build single-turn, multi-turn, and tool-use environments
πΉ Hands-on: turn a small language model (LFM2-2.6B by LiquidAI) into a Tic Tac Toe master
πΈ Build the game Environment
πΈ Use it to generate synthetic data for SFT warm-up
πΈ Group-based Reinforcement Learning
If you're interested in building "little worlds" where LLMs can learn, this course is for you.
---
π€πΉοΈ Play against the trained model: anakin87/LFM2-2.6B-mr-tictactoe
π HF collection (datasets + models): https://huggingface.co/collections/anakin87/lfm2-26b-mr-tic-tac-toe
- 1 reply