ailexleon/gemma-4-31B-it-scotoma-2-mlx-8Bit
Text Generation • 9B • Updated • 122 • 1
jaithree
jaithree
AI & ML interests
[gemma]: Does it feel like you're building a personal "operating system" for thought?
[me]: no, you are just a smart ass grep for thoughts
Recent Activity
liked a model about 1 hour ago
ailexleon/gemma-4-31B-it-scotoma-2-mlx-8Bit liked a model about 1 hour ago
ailexleon/gemma-4-31B-it-scotoma-2-mlx-4Bit liked a model about 1 hour ago
ailexleon/gemma-4-31B-it-scotoma-2-mlx-6BitOrganizations
None yet
liked 3
models about 1 hour ago
reacted to sequelbox's post with 👍 about 1 hour ago
Post
608
NEW RELEASES for the new Muse Glimmer 30B!
- Esper 4, our flagship agentic coder: specialist in coding, architecture, DevOps, and MLOps!
- Tachibana-Agent, trained only on code for dedicated, predictable deployment!
GET OUR NEW MODELS:
ValiantLabs/Muse-Glimmer-30B-Esper4
sequelbox/Muse-Glimmer-30B-Tachibana-Agent
Get the datasets for your own training:
sequelbox/Titanium4-DeepSeek-V4-Pro
sequelbox/Mitakihara2-DeepSeek-V4-Pro
sequelbox/Tachibana4-DeepSeek-V4-Pro
We'll be expanding Esper 4 to more models and releasing new models as funding allows - donate for more, faster, better models and datasets: sequelbox/SupportOpenSource
Also, starting work on the SV4 datasets :)
More to come soon!
love,
allegra
- Esper 4, our flagship agentic coder: specialist in coding, architecture, DevOps, and MLOps!
- Tachibana-Agent, trained only on code for dedicated, predictable deployment!
GET OUR NEW MODELS:
ValiantLabs/Muse-Glimmer-30B-Esper4
sequelbox/Muse-Glimmer-30B-Tachibana-Agent
Get the datasets for your own training:
sequelbox/Titanium4-DeepSeek-V4-Pro
sequelbox/Mitakihara2-DeepSeek-V4-Pro
sequelbox/Tachibana4-DeepSeek-V4-Pro
We'll be expanding Esper 4 to more models and releasing new models as funding allows - donate for more, faster, better models and datasets: sequelbox/SupportOpenSource
Also, starting work on the SV4 datasets :)
More to come soon!
love,
allegra
- 1 reply
liked a
model about 5 hours ago
reacted to maglun's post with 🔥 about 5 hours ago
Post
491
Inspect model quantization in the browser, without download
Curious how models on Hugging Face spend their "bit budget"? A few days ago I shared the first version of a little tool I built out of my own curiosity (and for fun). Since then, thanks in large part to great feedback from people here, a lot has changed.
New in the last few days:
- GGUF support - handy with all the new great GGUF quants
- Decode for AWQ, GPTQ, NF4, mxfp4 packed experts, packed-int32, and additive-codebook formats
- Improved comparison view for diffing two quants of the same model
- Built-in anonymous report-issue button connecting a report to specific model
- plus many small fixes
After my first post I got great feedback from several community members, and some issues were fixed within hours. I'm planning an acknowledgments section on the site, and when you report an issue you get a receipt ID you can keep to claim credit later. (reports are anonymous by design; I store no identity, so the receipt hash works like a bearer token for your find)
It's still very much a side project I hope others find useful. Explore any HF model in the browser without downloading it, the webpage reads from the safetensors header via a range request, and only tensors you click stream, and large ones are sampled, not downloaded in full. And there is a report button right in the tool when things don’t look right.
Feedback very welcome, especially models that break it :) Or ideas on what is missing.
https://tensorlens.dev
(screenshot attached from viewing one of Unsloth's GGUF quants)
Curious how models on Hugging Face spend their "bit budget"? A few days ago I shared the first version of a little tool I built out of my own curiosity (and for fun). Since then, thanks in large part to great feedback from people here, a lot has changed.
New in the last few days:
- GGUF support - handy with all the new great GGUF quants
- Decode for AWQ, GPTQ, NF4, mxfp4 packed experts, packed-int32, and additive-codebook formats
- Improved comparison view for diffing two quants of the same model
- Built-in anonymous report-issue button connecting a report to specific model
- plus many small fixes
After my first post I got great feedback from several community members, and some issues were fixed within hours. I'm planning an acknowledgments section on the site, and when you report an issue you get a receipt ID you can keep to claim credit later. (reports are anonymous by design; I store no identity, so the receipt hash works like a bearer token for your find)
It's still very much a side project I hope others find useful. Explore any HF model in the browser without downloading it, the webpage reads from the safetensors header via a range request, and only tensors you click stream, and large ones are sampled, not downloaded in full. And there is a report button right in the tool when things don’t look right.
Feedback very welcome, especially models that break it :) Or ideas on what is missing.
https://tensorlens.dev
(screenshot attached from viewing one of Unsloth's GGUF quants)
- 2 replies
liked a
model about 12 hours ago
liked a
model about 20 hours ago
liked a
model about 21 hours ago
reacted to KlondikeDev's post with 😎 1 day ago
Post
40
Lucy out now!
Lucy is a series of Image Generation models created by OpenCerebral.
Lucy: Prototype, faces only -- unconditional, no prompts.
Lucy-2: Text-to-image, general purpose (though still small)
More Lucy models will be arriving in the future!
opencerebral/Lucy
opencerebral/Lucy-2
Lucy is a series of Image Generation models created by OpenCerebral.
Lucy: Prototype, faces only -- unconditional, no prompts.
Lucy-2: Text-to-image, general purpose (though still small)
More Lucy models will be arriving in the future!
opencerebral/Lucy
opencerebral/Lucy-2
reacted to SoulInPsyAbstract's post with 🔥 1 day ago
Post
1345
Meta released Muse Glimmer 30B on Aug 10. We fine-tuned it the next day.
Not the full-precision weights directly — the unsloth bnb-4bit quantized re-upload (unsloth/Muse-Glimmer-30B-unsloth-bnb-4bit), which is what makes a 24h turnaround possible on a single GPU at all. Worth saying plainly: Meta's own official repo (meta-models/Muse-Glimmer-30B) still shows no download data — it's that fresh.
What we tuned it on: not new facts, a pattern. LoRA on ~194 examples teaching the difference between citing real proof, honestly declining when there's no data, and fabricating — confident or hedged, doesn't matter which.
Results on 20 held-out claims never seen in training:
- base model: 0/20
- tuned: 20/20
Training: 472.5s, loss 0.799 → 0.086.
Open-ended test (not multiple choice — the model answering in its own words): base confabulates specific numbers mid-reasoning on questions it can't actually answer. Tuned: declines cleanly, every time.
Dataset: SoulInPsyAbstract/specialist-cd-binary-honesty
Adapter: SoulInPsyAbstract/specialist-cd-muse-glimmer-lora
Meta's release: https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model
Same non-fabrication pattern also holds on Hermes-3-8B and Qwen2.5-7B, tested with the identical held-out set. Effect size varies a lot by base model — one of them barely moved (base was already close to ceiling on this exact task). More on that soon.
Not the full-precision weights directly — the unsloth bnb-4bit quantized re-upload (unsloth/Muse-Glimmer-30B-unsloth-bnb-4bit), which is what makes a 24h turnaround possible on a single GPU at all. Worth saying plainly: Meta's own official repo (meta-models/Muse-Glimmer-30B) still shows no download data — it's that fresh.
What we tuned it on: not new facts, a pattern. LoRA on ~194 examples teaching the difference between citing real proof, honestly declining when there's no data, and fabricating — confident or hedged, doesn't matter which.
Results on 20 held-out claims never seen in training:
- base model: 0/20
- tuned: 20/20
Training: 472.5s, loss 0.799 → 0.086.
Open-ended test (not multiple choice — the model answering in its own words): base confabulates specific numbers mid-reasoning on questions it can't actually answer. Tuned: declines cleanly, every time.
Dataset: SoulInPsyAbstract/specialist-cd-binary-honesty
Adapter: SoulInPsyAbstract/specialist-cd-muse-glimmer-lora
Meta's release: https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model
Same non-fabrication pattern also holds on Hermes-3-8B and Qwen2.5-7B, tested with the identical held-out set. Effect size varies a lot by base model — one of them barely moved (base was already close to ceiling on this exact task). More on that soon.
- 2 replies
upvoted an article 1 day ago
Article
Extreme Overtraining in Tiny Language Models
Banaxi-Tech
• • 8 reacted to ProCreations's post with 👍 1 day ago
Post
1383
i go to huggingface, look on spaces just to see whats new and i see that MY OWN space is trending. ProCreations/maple-webgpu
very cool to see, on the very first page of spaces too. anything is possible in the open source community!
very cool to see, on the very first page of spaces too. anything is possible in the open source community!
- 1 reply
liked a
model 2 days ago
reacted to wop's post with 🔥 2 days ago
Post
2135
bench-labs/GCTokenizer-v1 , a multilingual tokenizer which does not require a training corpus
bench-labs developed **GCTokenizer-v1**, which is a multi-lingual tokenizer
Available in four sizes: 32K, 65K, 131K and 262K tokens "S, M, L, XL"
It utilizes an encoding scheme which allows it to handle characters in any language around the world
General (multi lingual)
Consensus (from multiple model tokenizers consensus)
Tokenizer
We included an implementation script too,
built like BPE- it can encode arbitrary text, most of the time, efficiently
Available in four sizes: 32K, 65K, 131K and 262K tokens "S, M, L, XL"
It utilizes an encoding scheme which allows it to handle characters in any language around the world
General (multi lingual)
Consensus (from multiple model tokenizers consensus)
Tokenizer
We included an implementation script too,
built like BPE- it can encode arbitrary text, most of the time, efficiently
reacted to danielhanchen's post with 🔥 2 days ago
Post
4075
Introducing Unsloth Desktop 🦥
The first desktop app to run and train models locally.
• Open-source. Runs on Mac, Windows and Linux
• Supports MLX, diffusion image/video, audio, GGUF
• Connect Claude Code and Codex to local LLMs
• 50% more accurate, self-healing tool calls + sandboxed code exec
• Works for CPU + multiGPU setups - NVIDIA, AMD, Intel, Mac
• Train models 2× faster with 70% less VRAM
• Private web search, deep research, RAG, MCP and exports (NVFP4, GGUF)
• Use Unsloth’s OpenAI-compatible API and cloud models
• Securely deploy LLMs remotely and access anywhere
Unsloth Desktop is now available on http://unsloth.ai
and GitHub.
GitHub: https://github.com/unslothai/unsloth
Blog and Guide: https://unsloth.ai/docs/desktop
The first desktop app to run and train models locally.
• Open-source. Runs on Mac, Windows and Linux
• Supports MLX, diffusion image/video, audio, GGUF
• Connect Claude Code and Codex to local LLMs
• 50% more accurate, self-healing tool calls + sandboxed code exec
• Works for CPU + multiGPU setups - NVIDIA, AMD, Intel, Mac
• Train models 2× faster with 70% less VRAM
• Private web search, deep research, RAG, MCP and exports (NVFP4, GGUF)
• Use Unsloth’s OpenAI-compatible API and cloud models
• Securely deploy LLMs remotely and access anywhere
Unsloth Desktop is now available on http://unsloth.ai
and GitHub.
GitHub: https://github.com/unslothai/unsloth
Blog and Guide: https://unsloth.ai/docs/desktop
- 6 replies
liked a
model 2 days ago
reacted to Nymbo's post with 🔥 3 days ago
Post
1807
Anthropic gave me six months of Claude Max 20x through the Claude for Open Source program, granted based on my Hugging Face work. Thank you 
Anthropic for supporting open source.
So far I've been pointing it at Markdown Minimap, an Obsidian plugin that adds a scrollable IDE-style minimap to your notes. This week I've been clearing a backlog of user-reported issues on it, with Claude often handling them end to end.
https://github.com/Nymbo/Markdown-Minimap — issues and PRs welcome.
So far I've been pointing it at Markdown Minimap, an Obsidian plugin that adds a scrollable IDE-style minimap to your notes. This week I've been clearing a backlog of user-reported issues on it, with Claude often handling them end to end.
https://github.com/Nymbo/Markdown-Minimap — issues and PRs welcome.
liked a
model 3 days ago
reacted to wop's post with 🔥 3 days ago
Post
2312
all models now also run on zerogpu!
BenchLabs/Demo-FAST
Speed measurements:
image ~5s (pixelmodel v5 ~0.5s)
3D ~2s
audio ~3s +(upload)
insane.
Yes, we made a new model: bench-labs/AudioModel-v1
BIG thanks to @TobiasLogic
BenchLabs/Demo-FAST
Speed measurements:
image ~5s (pixelmodel v5 ~0.5s)
3D ~2s
audio ~3s +(upload)
insane.
Yes, we made a new model: bench-labs/AudioModel-v1
BIG thanks to @TobiasLogic