🔄 In a Training Loop

Thomas Yiu

legolasyiu

https://account.venmo.com/u/Thomas-Yiu

AI & ML interests

Superalignment , alignment, safety LLM, AI, AGI, ASI. Donation/Funding https://account.venmo.com/u/Thomas-Yiu

Recent Activity

updated a Space about 11 hours ago

EpistemeAI/ReasoningMedical27b

updated a model about 12 hours ago

EpistemeAI/Reasoning-Medical-27B

updated a model about 12 hours ago

EpistemeAI/Reasoning-Medical0.1-27B

View all activity

Organizations

posted an update 7 days ago

Post

150

Introducing Reasoning-Medical-27B is designed for advanced medical reasoning in professional medicine, medical genetics, college biology/medicine, and clinical knowledge. The model was fine-tuned on a large-scale dataset of 370,000 high-quality question-and-answer examples, incorporating Chain-of-Thought reasoning to improve step-by-step problem solving. Training was performed using the GRPO trainer with the Unsloth optimization method for efficient fine-tuning.
MedQA: 93% vs MedGemma 85.3%

Model: EpistemeAI/Reasoning-Medical-27B

# Benchmark

| Task              | Version | Filter         | n-shot | Metric      | Direction | Reasoning Medical 27B | Qwen 3.6 27B | MedGemma 1 27B |
|-------------------|--------:|----------------|-------:|-------------|:---------:|----------------------:|-------------:|----------------:|
| MMLU-Pro Biology  | 3.1     | custom-extract | 2      | exact_match | ↑         | 0.85                  | —            | —               |
| MMLU-ProX Biology | 0       | custom-extract | 2      | exact_match | ↑         | 0.80                  | —            | —               |
| MedQA             | YAML    | none           | 2      | acc         | ↑         | 0.93                  | 0.844        | 0.853           |

replied to their post 6 months ago

It uses recursive self improvement techniques. It also improves coding vs others in Humaneval.

replied to their post 6 months ago

Codeforce-GPT-oss-20b leads the benchmark, surpassing even larger models like Qwen 3 235B and DeepSeek-R1 70B. Its superior reasoning and code synthesis capabilities indicate an optimized training strategy rather than sheer scale dominance.

replied to their post 6 months ago

🧠 Model Benchmark Comparison

This table presents HumanEval benchmark scores across several large language models.

Model	HumanEval
Codeforce-GPT-oss-20b	90
Qwen 3 235B	80
DeepSeek-R1 70B	88
Phi-4 Reasoning	88
Llama 4 Scout	78
Llama 3.3 70B	83
Gemma 3 27B	76
GPT-OSS 20B	73
GPT-OSS 120B	71

posted an update 6 months ago

Post

335

We release open-weight early experimental Codeforce metatune-gpt20b, fine tuned version of OpenAI's gpt-oss-20b model, this is one of the first public release recursive self improving AI.

EpistemeAI/Codeforce-metatune-gpt20b

9 replies

replied to merve's post 9 months ago

Episteme released a Vibe Coding LLM with 20B params based on GPT OSS 20B, it is easier to Vibe Code. It is preview right now.
EpistemeAI/VibeCoder-20B-alpha

replied to clem's post over 1 year ago

Did you get a chance to try it?

replied to clem's post over 1 year ago

Hello Belo, nice to meet you, I am special fine tune models.
I have cool model, from EpistemeAI,
https://huggingface.co/EpistemeAI/Fireball-Meta-Llama-3.1-8B-Instruct-Agent-0.003-128K-code-ds-auto

reacted to KingNish's post with 👍 almost 2 years ago

Post

3649

Mistral Nemo is better than many models in 1st grader level reasoning.

reacted to davidberenstein1957's post with 🔥 almost 2 years ago

Post

3041

🚀 We will be generating a preference dataset for DPO/ORPO and cleaning it with AI feedback during our upcoming meetup!

In this session, we'll walk you through the essentials of building a distilabel pipeline by exploring two key use cases: cleaning an existing dataset and generating a preference dataset for DPO/ORPO. You’ll also learn how to make the most of AI feedback, integrating Argilla to gather human feedback and improve the overall data quality.

This session is perfect for you
- if you’re getting started with distilabel or synthetic data
- if you want to learn how to use LLM inference endpoints for **free**
- if you want to discover new functionalities
- if you want to provide us with new feedback

Sign up here: https://lu.ma/dt0c7jru

reacted to sequelbox's post with 👍 almost 2 years ago

Post

1617

NEW math-instruct model + dataset!

ValiantLabs/Llama3.1-8B-Cobalt is our new math-instruct model.
Trained using a synthetic math-instruct dataset generated with Llama 3.1 405b. Find the dataset here: sequelbox/Polytope

More to come soon :)

Thomas Yiu

AI & ML interests

Recent Activity

Organizations

legolasyiu's activity

🧠 Model Benchmark Comparison