DeepSeek-V3 on HF 

Reasoning and Coding benchmarks:

BigCodeBench-Hard:
- 🥇 1st Overall
- Complete: 40.5%
- Instruct: 28.4%
- Average: 34.5%
> Comparisons: Gemini-Exp-1206 (34.1%), o1-2024-12-17 (32.8%)

Aider Polyglot Leaderboard:
-🥈 2nd Place
- o1: 62%
- DeepSeek-V3 Preview: 48%
- Sonnet: 45%, Gemini-Exp-1206: 38%, o1-Mini: 33%

SWE-Bench Lite:
- Performance: 23.00%

Key Architecture Upgrades (V3 vs. V2):
> Vocab Size: 129,280 (↑ from 102,400)
> Hidden Size: 7,168 (↑ from 4,096)
> Layers: 61 (↑ from 30)
> Attention Heads: 128 (↑ from 32)
> Max Position Embeddings: 4,096 (↑ from 2,048)

DeepSeek-V3 represents a significant scale-up from its predecessor, with a massive 685B parameters—substantially larger than the largest LLaMA model (405B). It sets new standards in reasoning and coding, with extensive architectural and performance upgrades across all benchmarks.

don’t sleep on deepseek.

DeepSeek-V3 on HF 

Reasoning and Coding benchmarks:

BigCodeBench-Hard:
- 🥇 1st Overall
- Complete: 40.5%
- Instruct: 28.4%
- Average: 34.5%
> Comparisons: Gemini-Exp-1206 (34.1%), o1-2024-12-17 (32.8%)

Aider Polyglot Leaderboard:
-🥈 2nd Place
- o1: 62%
- DeepSeek-V3 Preview: 48%
- Sonnet: 45%, Gemini-Exp-1206: 38%, o1-Mini: 33%

SWE-Bench Lite:
- Performance: 23.00%

Key Architecture Upgrades (V3 vs. V2):
> Vocab Size: 129,280 (↑ from 102,400)
> Hidden Size: 7,168 (↑ from 4,096)
> Layers: 61 (↑ from 30)
> Attention Heads: 128 (↑ from 32)
> Max Position Embeddings: 4,096 (↑ from 2,048)

DeepSeek-V3 represents a significant scale-up from its predecessor, with a massive 685B parameters—substantially larger than the largest LLaMA model (405B). It sets new standards in reasoning and coding, with extensive architectural and performance upgrades across all benchmarks.

don’t sleep on deepseek.

https://huggingface.co/deepseek-ai/DeepSeek-V3-Base/tree/main

DeepSeek-V3 excels in reasoning and coding benchmarks with impressive architecture upgrades, setting new standards in the field. Highly recommended