42 14 84

Ivan Fioravanti PRO

ivanfioravanti

AI & ML interests

None yet

Recent Activity

upvoted an article 5 days ago

You could have designed state of the art positional encoding

reacted to wolfram's post with 🔥 14 days ago

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!

liked a model 14 days ago

mlx-community/Qwen3-30B-A3B-4bit-DWQ-0508

View all activity

Organizations

ivanfioravanti's activity

liked a model 14 days ago

mlx-community/Qwen3-30B-A3B-4bit-DWQ-0508

Text Generation • Updated 14 days ago • 480 • 13

liked a model 21 days ago

deepseek-ai/DeepSeek-Prover-V2-7B

Updated 22 days ago • 28.1k • 101

liked a model 24 days ago

Qwen/Qwen3-0.6B-FP8

Text Generation • Updated 1 day ago • 6.15k • 44

liked a model 27 days ago

cognitivecomputations/Dolphin3.0-Mistral-24B

Text Generation • Updated 27 days ago • 1.71k • 59

liked a dataset 28 days ago

nvidia/OpenMathReasoning

Viewer • Updated 13 days ago • 5.47M • 46k • 248

liked a model 29 days ago

nari-labs/Dia-1.6B

Text-to-Speech • Updated 8 days ago • 196k • • 2.33k

liked 6 models about 2 months ago

meta-llama/Llama-4-Maverick-17B-128E-Instruct

Image-Text-to-Text • Updated Apr 9 • 52.9k • • 331

meta-llama/Llama-4-Scout-17B-16E-Instruct

Image-Text-to-Text • Updated Apr 9 • 329k • • 913

nomic-ai/nomic-embed-text-v2-moe

liked a model 2 months ago

nvidia/Llama-3_3-Nemotron-Super-49B-v1

Text Generation • Updated 14 days ago • 210k • • 292

liked a dataset 2 months ago

glaiveai/reasoning-v1-20m

Viewer • Updated Mar 19 • 22.2M • 4.7k • 209

liked 2 models 2 months ago

mlx-community/Mistral-Small-24B-Instruct-2501-bf16

Updated Jan 30 • 42 • 6

mlx-community/Kokoro-82M-bf16

Text-to-Speech • Updated Mar 8 • 1.1k • 18

liked a model 3 months ago

Qwen/QwQ-32B

Text Generation • Updated Mar 11 • 347k • • 2.76k

liked a model 4 months ago

unsloth/DeepSeek-R1

Text Generation • Updated 29 days ago • 612 • 50

liked a dataset 4 months ago

cognitivecomputations/dolphin-r1

Viewer • Updated Jan 30 • 814k • 1.66k • 280

liked a Space 4 months ago

2.67k

Hunyuan3D-2.0

🌍

Text-to-3D and Image-to-3D Generation