Qwen/Qwen3.5-0.8B
Image-Text-to-Text ā¢ 0.9B ā¢ Updated ā¢ 3.67M ā¢ 807
alkinun
AtAndDev
AI & ML interests
decentralize
Recent Activity
liked a model 1 day ago
Qwen/Qwen3.5-0.8B liked a model 1 day ago
google/gemma-4-31B-itOrganizations
liked a
model 1 day ago
reacted to savoji's post with š„ 1 day ago
Post
3,831
LFM3-1.7B from @LiquidAI is the #1 trending model on the Hub (855 likes). It takes about 1 GB at 4-bit, and in the vendor's own table it beats Qwen3.5 9B on several agent benchmarks.
The edge decision moves from "are small models good enough?" to "which architecture and runtime do I pick?" One 1.7B backbone now ships as text, VL-2B and Omni-2B checkpoints with open weights.
š Architecture and speed
ā¢ 24 layers: 13 gated convolutions (kernel 7), 5 sliding-window attention (256 tokens), 6 global attention, chosen by a search over 14 device and backend targets
ā¢ llama.cpp Q4_0 at 32K context: 1.8 to 2.8x faster decode than the next model on 3 devices
ā¢ Full 128K context needs 1.6 GB of KV cache vs 4.3 GB (Qwen3.5-4B) and 5.6 GB (MiniCPM5-2B)
š Liquid AI's table
ā¢ Average 49.37 vs 47.24 for Qwen3.5 9B; BFCL V4 71.39 vs 62.83; WebArena-Verified 38.42 vs 28.94
ā¢ Qwen3.5 9B still wins SWE-bench Verified (51.20 vs 43.40) and GPQA Diamond (80.40 vs 63.37)
š ļø Try it
ā¢ GGUF (174,177 downloads last month) runs in llama.cpp b11960+, Ollama and LM Studio; MLX 4-bit for Apple Silicon; SGLang for GPUs
ā¢ llama-cli -hf LiquidAI/LFM3-1.7B-GGUF:Q4_K_M --temp 0.1 --top-k 50 --repeat-penalty 1.1
ā¢ Omni-2B runs in the browser in the WebGPU Space
ā ļø Caveats
ā¢ The numbers are vendor-reported, with no independent reproduction in the sources. The card says it is not for agentic coding or complex multi-step reasoning.
LiquidAI/LFM3-1.7B
LiquidAI/LFM3-1.7B-GGUF
LiquidAI/LFM3-Omni-2B
LiquidAI/LFM3-Omni-2B-WebGPU
Has a sub-2B model held up on your own tool-calling harness?
ā¢ Medium: https://medium.com/p/05507e4e81c8
ā¢ Substack: https://sajjadpakdamansavoji.substack.com/p/liquid-ais-lfm3-turns-the-small-model
ā¢ X: https://x.com/saavoji/status/2117307514177941672
ā¢ Website: https://sajjadpsavoji.github.io/blog#liquid-ais-lfm3
The edge decision moves from "are small models good enough?" to "which architecture and runtime do I pick?" One 1.7B backbone now ships as text, VL-2B and Omni-2B checkpoints with open weights.
š Architecture and speed
ā¢ 24 layers: 13 gated convolutions (kernel 7), 5 sliding-window attention (256 tokens), 6 global attention, chosen by a search over 14 device and backend targets
ā¢ llama.cpp Q4_0 at 32K context: 1.8 to 2.8x faster decode than the next model on 3 devices
ā¢ Full 128K context needs 1.6 GB of KV cache vs 4.3 GB (Qwen3.5-4B) and 5.6 GB (MiniCPM5-2B)
š Liquid AI's table
ā¢ Average 49.37 vs 47.24 for Qwen3.5 9B; BFCL V4 71.39 vs 62.83; WebArena-Verified 38.42 vs 28.94
ā¢ Qwen3.5 9B still wins SWE-bench Verified (51.20 vs 43.40) and GPQA Diamond (80.40 vs 63.37)
š ļø Try it
ā¢ GGUF (174,177 downloads last month) runs in llama.cpp b11960+, Ollama and LM Studio; MLX 4-bit for Apple Silicon; SGLang for GPUs
ā¢ llama-cli -hf LiquidAI/LFM3-1.7B-GGUF:Q4_K_M --temp 0.1 --top-k 50 --repeat-penalty 1.1
ā¢ Omni-2B runs in the browser in the WebGPU Space
ā ļø Caveats
ā¢ The numbers are vendor-reported, with no independent reproduction in the sources. The card says it is not for agentic coding or complex multi-step reasoning.
LiquidAI/LFM3-1.7B
LiquidAI/LFM3-1.7B-GGUF
LiquidAI/LFM3-Omni-2B
LiquidAI/LFM3-Omni-2B-WebGPU
Has a sub-2B model held up on your own tool-calling harness?
ā¢ Medium: https://medium.com/p/05507e4e81c8
ā¢ Substack: https://sajjadpakdamansavoji.substack.com/p/liquid-ais-lfm3-turns-the-small-model
ā¢ X: https://x.com/saavoji/status/2117307514177941672
ā¢ Website: https://sajjadpsavoji.github.io/blog#liquid-ais-lfm3
- 10 replies
liked 2
models 1 day ago
liked 3
models 4 days ago
reacted to sergiopaniego's post with ššš„ 4 days ago
Post
3,040
Mistral Large 4 (Le Chonk š) is now live on the Hub and you can post-train it with TRL today!
SFT + LoRA example ready, runs on a single 8xB300 node
> model: mistralai/Mistral-Large-4-1T-A52B
> TRL example: https://github.com/huggingface/trl/tree/main/examples/sft_mistral_large_4
> blog: https://huggingface.co/blog/mistral-large4
SFT + LoRA example ready, runs on a single 8xB300 node
> model: mistralai/Mistral-Large-4-1T-A52B
> TRL example: https://github.com/huggingface/trl/tree/main/examples/sft_mistral_large_4
> blog: https://huggingface.co/blog/mistral-large4
liked a
model 4 days ago
liked a
model 5 days ago
liked a
model 6 days ago
liked 2
models 8 days ago
replied to Banaxi-Tech's post 9 days ago
gemini 2.0 flash was peak, went downhill after that.
reacted to mithulaartigala's post with š„ 10 days ago
Post
3,271
We need to make software where people can start using decision models.
reacted to appvoid's post with š„ 10 days ago
Post
3,995
You know, I can't believe we are actually doing it, literally advancing the frontier at the edge! We've come a long way since the first models on open slm leaderboard, and, as many things that change, performance is one of the main things that has exploded.
7 years ago, the best we had was GPT-2. Today, we got 90m models beating smollm2! I can't barely see what a future with increasingly powerful small models will look like.
All I can say is that the best is yet to come. And will be agentic.
7 years ago, the best we had was GPT-2. Today, we got 90m models beating smollm2! I can't barely see what a future with increasingly powerful small models will look like.
All I can say is that the best is yet to come. And will be agentic.
- 11 replies
liked a
model 10 days ago
replied to Banaxi-Tech's post 10 days ago
im talking about gemini, gemma always cooks. gemini is shit always.