49
Pixel Reasoner
⚡
The demo for pixel reasoner
Bertrand Chevrier
kramp
AI & ML interests
text 2 speech, ai for music writting
Recent Activity
liked
a Space
7 days ago
TIGER-Lab/Pixel-Reasoner
upvoted
a
changelog
12 days ago
AI-generated Abstract summaries on Hugging Face Papers
upvoted
a
changelog
13 days ago
Filter by MCP compatibility available in HF Spaces
Organizations
kramp's activity
liked
a
Space
7 days ago
upvoted
a
changelog
12 days ago
Changelog
AI-generated Abstract summaries on Hugging Face Papers
•
65
upvoted
a
changelog
13 days ago
Changelog
Filter by MCP compatibility available in HF Spaces
•
70
liked
a
model
14 days ago
upvoted
2
articles
14 days ago
Article
The Transformers Library: standardizing model definitions
By
and 3 others
•
•
109
Article
Microsoft and Hugging Face expand collaboration
By
and 2 others
•
•
20
reacted to
AdinaY's
post with 🔥
14 days ago
Post
2396
Dolphin 🔥 A multimodal document image parsing model from ByteDance
, built on an analyze-then-parse paradigm.
ByteDance/Dolphin
✨ MIT licensed
✨ Handles text, tables, figures & formulas via:
- Reading-order layout analysis
- Parallel parsing with smart prompts
, built on an analyze-then-parse paradigm.
ByteDance/Dolphin
✨ MIT licensed
✨ Handles text, tables, figures & formulas via:
- Reading-order layout analysis
- Parallel parsing with smart prompts
liked
a
model
15 days ago
liked
a
Space
19 days ago
liked
a
model
19 days ago
liked
a
Space
21 days ago
liked
a
Space
23 days ago
liked
a
model
26 days ago
upvoted
an
article
26 days ago
Article
AI Personas: The Impact of Design Choices
By
and 1 other
•
•
14
reacted to
wolfram's
post with 👍
26 days ago
Post
7166
Finally finished my extensive **Qwen 3 evaluations** across a range of formats and quantisations, focusing on **MMLU-Pro** (Computer Science).
A few take-aways stood out - especially for those interested in local deployment and performance trade-offs:
1️⃣ **Qwen3-235B-A22B** (via Fireworks API) tops the table at **83.66%** with ~55 tok/s.
2️⃣ But the **30B-A3B Unsloth** quant delivered **82.20%** while running locally at ~45 tok/s and with zero API spend.
3️⃣ The same Unsloth build is ~5x faster than Qwen's **Qwen3-32B**, which scores **82.20%** as well yet crawls at <10 tok/s.
4️⃣ On Apple silicon, the **30B MLX** port hits **79.51%** while sustaining ~64 tok/s - arguably today's best speed/quality trade-off for Mac setups.
5️⃣ The **0.6B** micro-model races above 180 tok/s but tops out at **37.56%** - that's why it's not even on the graph (50 % performance cut-off).
All local runs were done with LM Studio on an M4 MacBook Pro, using Qwen's official recommended settings.
**Conclusion:** Quantised 30B models now get you ~98 % of frontier-class accuracy - at a fraction of the latency, cost, and energy. For most local RAG or agent workloads, they're not just good enough - they're the new default.
Well done, Qwen - you really whipped the llama's ass! And to OpenAI: for your upcoming open model, please make it MoE, with toggleable reasoning, and release it in many sizes. *This* is the future!
A few take-aways stood out - especially for those interested in local deployment and performance trade-offs:
1️⃣ **Qwen3-235B-A22B** (via Fireworks API) tops the table at **83.66%** with ~55 tok/s.
2️⃣ But the **30B-A3B Unsloth** quant delivered **82.20%** while running locally at ~45 tok/s and with zero API spend.
3️⃣ The same Unsloth build is ~5x faster than Qwen's **Qwen3-32B**, which scores **82.20%** as well yet crawls at <10 tok/s.
4️⃣ On Apple silicon, the **30B MLX** port hits **79.51%** while sustaining ~64 tok/s - arguably today's best speed/quality trade-off for Mac setups.
5️⃣ The **0.6B** micro-model races above 180 tok/s but tops out at **37.56%** - that's why it's not even on the graph (50 % performance cut-off).
All local runs were done with LM Studio on an M4 MacBook Pro, using Qwen's official recommended settings.
**Conclusion:** Quantised 30B models now get you ~98 % of frontier-class accuracy - at a fraction of the latency, cost, and energy. For most local RAG or agent workloads, they're not just good enough - they're the new default.
Well done, Qwen - you really whipped the llama's ass! And to OpenAI: for your upcoming open model, please make it MoE, with toggleable reasoning, and release it in many sizes. *This* is the future!
- 4 replies
liked
a
dataset
29 days ago
updated
a
collection
30 days ago
liked
2
models
about 1 month ago
liked
a
Space
about 1 month ago