Warning: file_put_contents(aCache/aDaily/post/opendatascience/--): Failed to open stream: No space left on device in /var/www/tg-me/post.php on line 50
Data Science by ODS.ai 🦜 | Telegram Webview: opendatascience/2251 -
Telegram Group & Telegram Channel
Telegram Group » Korea » Data Science by ODS.ai 🦜 » Telegram Webview » Post 2251
Data Science by ODS.ai 🦜
🌟 DeepCoder-14B

New code reasoning LLM fine-tuned from DeepSeek-R1-Distill-Qwen-14B using distributed RL with GRPO+ and iterative context lengthening. Trained on ~24K coding problems (TACO-Verified, PrimeIntellect SYNTHETIC-1, LCB v5), it improves Pass@1 on LiveCodeBench v5 to 60.6%, +7.6% over base and on par with OpenAI o3-mini.

- GRPO+: removes KL/entropy loss for stability; adds offline difficulty filtering, DAPO-inspired loss masking, and reward clipping.
- Iterative context scaling: 16K→32K→64K generalization with improved long-context reasoning.

Eval: Strong results on LiveCodeBench, Codeforces, HumanEval+

Open weights🔥

https://huggingface.co/agentica-org/DeepCoder-14B-Preview

@opendatascience
Please open Telegram to view this post
VIEW IN TELEGRAM
www.tg-me.com/kr/Data Science by ODS ai 🦜/com.opendatascience/2251
3.1K views



tg-me.com/opendatascience/2251
Create:
Last Update:

🌟 DeepCoder-14B

New code reasoning LLM fine-tuned from DeepSeek-R1-Distill-Qwen-14B using distributed RL with GRPO+ and iterative context lengthening. Trained on ~24K coding problems (TACO-Verified, PrimeIntellect SYNTHETIC-1, LCB v5), it improves Pass@1 on LiveCodeBench v5 to 60.6%, +7.6% over base and on par with OpenAI o3-mini.

- GRPO+: removes KL/entropy loss for stability; adds offline difficulty filtering, DAPO-inspired loss masking, and reward clipping.
- Iterative context scaling: 16K→32K→64K generalization with improved long-context reasoning.

Eval: Strong results on LiveCodeBench, Codeforces, HumanEval+

Open weights🔥

https://huggingface.co/agentica-org/DeepCoder-14B-Preview

@opendatascience

BY Data Science by ODS.ai 🦜




Share with your friend now:
tg-me.com/opendatascience/2251

View MORE
Open in Telegram


Data Science by ODS ai 🦜 Telegram | DID YOU KNOW?

Date: |

Telegram announces Anonymous Admins

The cloud-based messaging platform is also adding Anonymous Group Admins feature. As per Telegram, this feature is being introduced for safer protests. As per the Telegram blog post, users can “Toggle Remain Anonymous in Admin rights to enable Batman mode. The anonymized admin will be hidden in the list of group members, and their messages in the chat will be signed with the group name, similar to channel posts.”

Data Science by ODS ai 🦜 from kr



Warning: filemtime(): stat failed for aCache/aDaily/post/opendatascience/-- in /var/www/tg-me/post.php on line 333

Warning: filemtime(): stat failed for aCache/aDaily/post/opendatascience/-- in /var/www/tg-me/post.php on line 334

🌟 DeepCoder-14BNew code reasoning LLM fine-tuned from DeepSeek-R1-Distill-Qwen-14B using distributed RL with GRPO+ and iterative context lengthening. Trained on ~24K coding problems (TACO-Verified

 Data Science by ODS.ai 🦜 TG
Webview: 2251
Data Science by ODS.ai 🦜.Telegram Webview
Data Science by ODS.ai 🦜 Telegram TG Channel
Telegram Updated:
Telegram Data Science by ODS.ai 🦜
FROM USA