Post
4,068
We are working on LFM3-1.7B-Ox, a compact model for repo-level programming, agentic coding, and tinkering.
- 2 replies
alkinun
AtAndDev
AI & ML interests
decentralize
Recent Activity
reacted to DedeProGames's post with π₯ about 3 hours ago
We are working on LFM3-1.7B-Ox, a compact model for repo-level programming, agentic coding, and tinkering. reacted to Banaxi-Tech's post with π₯ about 14 hours ago
We're introducing KV1 1.2B.
Its our first KV1 model using a 2 bit KV cache, based on LFM 2.5 1.2B.
KV1 1.2B retains 75.8% of original intelligence.
More details: https://huggingface.co/spaces/klein-ml/blog#kv1-1.2b
Model https://huggingface.co/klein-ml/KV1-1.2B
reacted to Banaxi-Tech's post with π₯ 1 day ago
BananaMind 3 is coming.
Biggest context in all SLMs with NSA.
10M, 50M, 150M, 400M.
And maybe, 1B.Organizations
reacted to DedeProGames's post with π₯ about 3 hours ago
reacted to Banaxi-Tech's post with π₯ about 14 hours ago
Post
2,170
We're introducing KV1 1.2B.
Its our first KV1 model using a 2 bit KV cache, based on LFM 2.5 1.2B.
KV1 1.2B retains 75.8% of original intelligence.
More details: klein-ml/blog
Model klein-ml/KV1-1.2B
Its our first KV1 model using a 2 bit KV cache, based on LFM 2.5 1.2B.
KV1 1.2B retains 75.8% of original intelligence.
More details: klein-ml/blog
Model klein-ml/KV1-1.2B
- 2 replies
reacted to Banaxi-Tech's post with π₯ 1 day ago
Post
6,071
BananaMind 3 is coming.
Biggest context in all SLMs with NSA.
10M, 50M, 150M, 400M.
And maybe, 1B.
Biggest context in all SLMs with NSA.
10M, 50M, 150M, 400M.
And maybe, 1B.
- 14 replies
reacted to SeaWolf-AI's post with π₯π 1 day ago
Post
6,498
π¦οΈ Which weather model is most accurate? We score them every day, in the open.
We released two things on Hugging Face:
π Global Weather Model Leaderboard
Every day it records what each global model predicted for 24 world cities, compares it with what actually happened (ERA5), and updates the ranking. ECMWF IFS and AIFS, NOAA GFS and AIGFS, DWD ICON, JMA, ECCC GEM, MET Norway, side by side, on the same cases, since 1 January 2026.
Four elements, four tabs: max temperature, min temperature, wind, precipitation. Forecast leads from 1 to 7 days, with 8 to 10 days joining as they are verified.
𧬠Darwin-Prediction-Weather
VIDRAFT's weather prediction system: a fusion engine that combines global forecasts, learning only from days already observed. It holds the lowest error in all four elements at every lead on the 2026 record, and the lowest error in all 15 evaluated cells of the WeatherBench 2 (2020) protocol.
What the 2026 record shows so far:
β’ AI models now lead physics models at the major centres (AIFS vs IFS, AIGFS vs GFS)
β’ Fusion goes further: 7-day max temperature RMSE 2.09 Β°C vs 2.41 Β°C for the next model
Also inside: yesterday's results with a daily WIN for the closest model, a 10-day outlook for every city, champions by city and by month, and a Live Earth map with wind, clouds, radar and temperature.
Every forecast and every observed value is open as CSV, so anyone can check the scores.
π Article: https://huggingface.co/blog/FINAL-Bench/which-weather-model-is-most-accurate-we-score-them
π Leaderboard: FINAL-Bench/Weather-Model-Leaderboard
𧬠Model: FINAL-Bench/Darwin-Prediction-Weather
π Daily record: FINAL-Bench/Weather-Model-Forecast-Archive
Questions, corrections and new models to add are welcome.
We released two things on Hugging Face:
π Global Weather Model Leaderboard
Every day it records what each global model predicted for 24 world cities, compares it with what actually happened (ERA5), and updates the ranking. ECMWF IFS and AIFS, NOAA GFS and AIGFS, DWD ICON, JMA, ECCC GEM, MET Norway, side by side, on the same cases, since 1 January 2026.
Four elements, four tabs: max temperature, min temperature, wind, precipitation. Forecast leads from 1 to 7 days, with 8 to 10 days joining as they are verified.
𧬠Darwin-Prediction-Weather
VIDRAFT's weather prediction system: a fusion engine that combines global forecasts, learning only from days already observed. It holds the lowest error in all four elements at every lead on the 2026 record, and the lowest error in all 15 evaluated cells of the WeatherBench 2 (2020) protocol.
What the 2026 record shows so far:
β’ AI models now lead physics models at the major centres (AIFS vs IFS, AIGFS vs GFS)
β’ Fusion goes further: 7-day max temperature RMSE 2.09 Β°C vs 2.41 Β°C for the next model
Also inside: yesterday's results with a daily WIN for the closest model, a 10-day outlook for every city, champions by city and by month, and a Live Earth map with wind, clouds, radar and temperature.
Every forecast and every observed value is open as CSV, so anyone can check the scores.
π Article: https://huggingface.co/blog/FINAL-Bench/which-weather-model-is-most-accurate-we-score-them
π Leaderboard: FINAL-Bench/Weather-Model-Leaderboard
𧬠Model: FINAL-Bench/Darwin-Prediction-Weather
π Daily record: FINAL-Bench/Weather-Model-Forecast-Archive
Questions, corrections and new models to add are welcome.
reacted to sergiopaniego's post with ππ₯ 5 days ago
Post
6,791
New post-training example: SFT on agent traces with TRL
β A large model works as a coding agent in OpenCode, run through OpenEnv
β The tasks it solves become the training data
β A 2B model goes from 5.5% to 34.4% on held-out tasks
About 500 traces get most of the gain. Five times more add only 3 to 4 points, so SFT gets the agent working and RL has to do the rest
Full guide, with the models, traces and every rollout open:
https://github.com/huggingface/trl/tree/main/examples/sft_agent_traces
β A large model works as a coding agent in OpenCode, run through OpenEnv
β The tasks it solves become the training data
β A 2B model goes from 5.5% to 34.4% on held-out tasks
About 500 traces get most of the gain. Five times more add only 3 to 4 points, so SFT gets the agent working and RL has to do the rest
Full guide, with the models, traces and every rollout open:
https://github.com/huggingface/trl/tree/main/examples/sft_agent_traces
- 1 reply
replied to Banaxi-Tech's post 6 days ago
oh nice! well, good luck. very excited.
replied to Banaxi-Tech's post 6 days ago
cool, but what happened to the promise? :)
reacted to Banaxi-Tech's post with π₯ 6 days ago
Post
4,902
We're excited to release BananaMind 2.5 Pro.
It's done some post-training on our BananaMind 2 Pro base, via distillation of SmolLM2 360M.
It scores 28.98 INT INDEX compared to 24.96 of BananaMind 2 Pro on the AxiomicLabs/Open_SLM_Leaderboard
Check it out at BananaMind/BananaMind-2.5-Pro
BananaMind
@ArushCodes
@Bc-AI
It's done some post-training on our BananaMind 2 Pro base, via distillation of SmolLM2 360M.
It scores 28.98 INT INDEX compared to 24.96 of BananaMind 2 Pro on the AxiomicLabs/Open_SLM_Leaderboard
Check it out at BananaMind/BananaMind-2.5-Pro
@ArushCodes
@Bc-AI
- 53 replies
reacted to savoji's post with π₯ 7 days ago
Post
2,512
Idol, a new Apache-2.0 image-to-3D model from @speridlabs , makes a 3D Gaussian splat from one image in 7.2 GB of peak VRAM. It is the #1 trending model on the Hub with 1,057 likes.
It writes a standard 3DGS .ply with up to 1,048,576 Gaussians, including sides the picture never showed, and does background removal and framing itself. The catch: the only quality evidence is the authors' own benchmark, on a repo created 2026-10-28.
π The numbers
β’ Authors' benchmark, 1 x H100: best in 6 of 8 columns vs TripoSplat, TRELLIS and LGM (SSIM 0.726, LPIPS 0.213, DINOv2 0.781, CLIP 0.885)
β’ LGM wins PSNR (12.46 vs 12.25) and speed (1.1 s). Idol takes 6.8 s vs 7.1 s for TRELLIS and 9.9 s for TripoSplat
β’ 262,144 Gaussians, 50 steps: 3.4 s on B300, 6.7 s on H100, 19.1 s on RTX 4090, 46.2 s on a 10 GB RTX 3080
π οΈ Try it
β’ Install idol-sdk, accept the terms of the gated DINOv3 and FLUX.2-dev models, run hf auth login
β’ python -m idol.sample --image photo.jpg --mode real
β’ --num_gaussians runs from 32,768 (0.4 MB) to 1,048,576 (12.2 MB)
β οΈ Caveats
β’ Test set and size are not named, and no independent evaluation exists yet
β’ The card says "under 2 seconds", but the timing table shows 3.4 s at best
β’ Single objects only, NVIDIA CUDA required, no failure cases shown
speridlabs/idol
https://speridlabs.com/research/idol
If you run it on your own messy photos, how do the unseen sides hold up?
π Read it elsewhere
β’ The deep dive on Medium: https://medium.com/p/5f660ae6ae56
β’ The long read on Substack: https://sajjadpakdamansavoji.substack.com/p/speridlabs-idol-builds-3d-splats
β’ The thread on X: https://x.com/saavoji/status/2119137739580858545
β’ Every version on sajjadpsavoji.github.io: https://sajjadpsavoji.github.io/blog#speridlabs-idol
It writes a standard 3DGS .ply with up to 1,048,576 Gaussians, including sides the picture never showed, and does background removal and framing itself. The catch: the only quality evidence is the authors' own benchmark, on a repo created 2026-10-28.
π The numbers
β’ Authors' benchmark, 1 x H100: best in 6 of 8 columns vs TripoSplat, TRELLIS and LGM (SSIM 0.726, LPIPS 0.213, DINOv2 0.781, CLIP 0.885)
β’ LGM wins PSNR (12.46 vs 12.25) and speed (1.1 s). Idol takes 6.8 s vs 7.1 s for TRELLIS and 9.9 s for TripoSplat
β’ 262,144 Gaussians, 50 steps: 3.4 s on B300, 6.7 s on H100, 19.1 s on RTX 4090, 46.2 s on a 10 GB RTX 3080
π οΈ Try it
β’ Install idol-sdk, accept the terms of the gated DINOv3 and FLUX.2-dev models, run hf auth login
β’ python -m idol.sample --image photo.jpg --mode real
β’ --num_gaussians runs from 32,768 (0.4 MB) to 1,048,576 (12.2 MB)
β οΈ Caveats
β’ Test set and size are not named, and no independent evaluation exists yet
β’ The card says "under 2 seconds", but the timing table shows 3.4 s at best
β’ Single objects only, NVIDIA CUDA required, no failure cases shown
speridlabs/idol
https://speridlabs.com/research/idol
If you run it on your own messy photos, how do the unseen sides hold up?
π Read it elsewhere
β’ The deep dive on Medium: https://medium.com/p/5f660ae6ae56
β’ The long read on Substack: https://sajjadpakdamansavoji.substack.com/p/speridlabs-idol-builds-3d-splats
β’ The thread on X: https://x.com/saavoji/status/2119137739580858545
β’ Every version on sajjadpsavoji.github.io: https://sajjadpsavoji.github.io/blog#speridlabs-idol
- 4 replies
reacted to danielhanchen's post with β€οΈππ₯ 7 days ago
Post
6,025
Qwen3.8-Flash-Next GGUFs now run faster with less thinking and less token use.
Unsloth Dynamic V3.1 GGUFs bring thinking & token use to match 95% of full precision Qwen3.8-Flash.
Results: 50% fewer tokens + 15% higher accuracy vs same size quants.
GGUF: unsloth/Qwen3.8-Flash-Next-GGUF
Analysis/Guide: https://unsloth.ai/docs/models/qwen3.8-next
Unsloth Dynamic V3.1 GGUFs bring thinking & token use to match 95% of full precision Qwen3.8-Flash.
Results: 50% fewer tokens + 15% higher accuracy vs same size quants.
GGUF: unsloth/Qwen3.8-Flash-Next-GGUF
Analysis/Guide: https://unsloth.ai/docs/models/qwen3.8-next
- 3 replies
reacted to SeaWolf-AI's post with π₯ 8 days ago
Post
8,254
SuperCLUEβs November 2026 results highlight why category scores matter when choosing a model.
Across the 19 published entries, here are the highest displayed scores in each category, followed by the leader among models labeled open:
β’ Agentic coding: GPT-6.1-Sol (69.23); open: DeepSeek-V4.1-Flash-1017 (65.38).
β’ Mathematical and scientific reasoning: GPT-6.1-Sol (76.32); open: Darwin-180B-RSI-R3 (73.68).
β’ Hallucination control: Gemini-4-Argon (83.19); open: Kimi-K3.1-Pro (67.78).
β’ Precise instruction following: Gemini-4-Argon and GPT-6.1-Sol (68.97 each); open: Kimi-K3.1-Pro (44.83).
β’ Ultra-long context: Gemini-4-Argon (95.63); open: Kimi-K3.1-Pro (87.25).
On the overall leaderboard, Gemini-4-Argon (73.62) and GPT-6.1-Sol (71.76) have the highest displayed scores, but neither has an official rank number. Kimi-K3.1-Pro (61.46) holds official rank No. 1 and leads the open models overall, followed by DeepSeek-V4.1-Flash-1017 (58.95, official No. 2) and GLM-5.3 (50.66, official No. 3).
These comparisons include entries whose official rank is β-β. Among officially ranked open models only, DeepSeek leads mathematical and scientific reasoning at 71.05. βOpenβ follows SuperCLUEβs classification, not an independent license review. Scores are benchmark points, not error rates; small differences and results from a Chinese-language-focused evaluation need validation on your own workloads. Detailed evaluation settings and the aggregation formula were not verified for this summary.
Read the full breakdown: https://huggingface.co/blog/FINAL-Bench/superclue-november-2026-category-leaders-open-mode
Source: SuperCLUE results released November 5, 2026
https://www.superclueai.com/homepage
Official November data: https://www.superclueai.com/data/generalboard/2026%E5%B9%B411%E6%9C%88.xlsx
Across the 19 published entries, here are the highest displayed scores in each category, followed by the leader among models labeled open:
β’ Agentic coding: GPT-6.1-Sol (69.23); open: DeepSeek-V4.1-Flash-1017 (65.38).
β’ Mathematical and scientific reasoning: GPT-6.1-Sol (76.32); open: Darwin-180B-RSI-R3 (73.68).
β’ Hallucination control: Gemini-4-Argon (83.19); open: Kimi-K3.1-Pro (67.78).
β’ Precise instruction following: Gemini-4-Argon and GPT-6.1-Sol (68.97 each); open: Kimi-K3.1-Pro (44.83).
β’ Ultra-long context: Gemini-4-Argon (95.63); open: Kimi-K3.1-Pro (87.25).
On the overall leaderboard, Gemini-4-Argon (73.62) and GPT-6.1-Sol (71.76) have the highest displayed scores, but neither has an official rank number. Kimi-K3.1-Pro (61.46) holds official rank No. 1 and leads the open models overall, followed by DeepSeek-V4.1-Flash-1017 (58.95, official No. 2) and GLM-5.3 (50.66, official No. 3).
These comparisons include entries whose official rank is β-β. Among officially ranked open models only, DeepSeek leads mathematical and scientific reasoning at 71.05. βOpenβ follows SuperCLUEβs classification, not an independent license review. Scores are benchmark points, not error rates; small differences and results from a Chinese-language-focused evaluation need validation on your own workloads. Detailed evaluation settings and the aggregation formula were not verified for this summary.
Read the full breakdown: https://huggingface.co/blog/FINAL-Bench/superclue-november-2026-category-leaders-open-mode
Source: SuperCLUE results released November 5, 2026
https://www.superclueai.com/homepage
Official November data: https://www.superclueai.com/data/generalboard/2026%E5%B9%B411%E6%9C%88.xlsx
reacted to Quazim0t0's post with π₯ 9 days ago
Post
7,487
Quazim0t0/Qwenx86-4B-compact
βΆοΈ See it in action: Quazim0t0/Qwenx86-Debugger . Walk through actual recorded tool calls: the conversation, x86 registers and flags, code at the program counter, RAM and machine output, with all verified blocks on a timeline.
Qwen3.5-4B with a 64-bit x86 computer compiled into its weights: 13 transformer layers ahead of Qwen's own. There are no learned weights in them; all of their weights are computed by a compiler. Qwen invokes the computer through its own tool calls (<tool_call>), and the computer in the weights executes the tool. Calculator, Python, C compiler, notes store, physics/chemistry lookup and Python-docs lookup are all firmware in the checkpoint. This is the compact checkpoint: ~9 GB GPU memory, so it fits on a 16 GB GPU.
Edit: GGUF Release: Quazim0t0/Qwenx86-4B-GGUF
Credits: Qwenputer by Divyansh Agrawal (@divagr1925 , github.com/divagr18/qwenputer, MIT)
βΆοΈ See it in action: Quazim0t0/Qwenx86-Debugger . Walk through actual recorded tool calls: the conversation, x86 registers and flags, code at the program counter, RAM and machine output, with all verified blocks on a timeline.
Qwen3.5-4B with a 64-bit x86 computer compiled into its weights: 13 transformer layers ahead of Qwen's own. There are no learned weights in them; all of their weights are computed by a compiler. Qwen invokes the computer through its own tool calls (<tool_call>), and the computer in the weights executes the tool. Calculator, Python, C compiler, notes store, physics/chemistry lookup and Python-docs lookup are all firmware in the checkpoint. This is the compact checkpoint: ~9 GB GPU memory, so it fits on a 16 GB GPU.
Edit: GGUF Release: Quazim0t0/Qwenx86-4B-GGUF
Credits: Qwenputer by Divyansh Agrawal (@divagr1925 , github.com/divagr18/qwenputer, MIT)
- 1 reply
reacted to appvoid's post with π₯πβ€οΈ 9 days ago
Post
7,608
New leaderboard coming to the small language model community!
The leaderboard measures cognitive skills, which is more challenging and beyond language modeling and instruction-following. This is a first step towards more everyday useful and general small models.
dotlabs/cognirank
The leaderboard measures cognitive skills, which is more challenging and beyond language modeling and instruction-following. This is a first step towards more everyday useful and general small models.
dotlabs/cognirank
- 2 replies
reacted to danielhanchen's post with π 10 days ago
Post
7,417
Weβre releasing new Qwen3.8-Flash-Next GGUFs with 15% higher accuracy.
Unsloth Dynamic V3.1 outperforms others by >15% on Terminal Bench, DeepSWE & more benchmarks (256K context).
Run on 32GB+ RAM setups.
GGUF: unsloth/Qwen3.8-Flash-Next-GGUF
Blog: https://unsloth.ai/docs/models/qwen3.8-next
Unsloth Dynamic V3.1 outperforms others by >15% on Terminal Bench, DeepSWE & more benchmarks (256K context).
Run on 32GB+ RAM setups.
GGUF: unsloth/Qwen3.8-Flash-Next-GGUF
Blog: https://unsloth.ai/docs/models/qwen3.8-next