ukisai/Swift-Qwen3.8-27b
Image-Text-to-Text • 28B • Updated • 2.75k • 346
Mike Ravkine PRO
mike-ravkine
AI & ML interests
LLM Research / Development / Evaluation
Recent Activity
liked a model 5 days ago
ukisai/Swift-Qwen3.8-27b liked a model 6 days ago
IFM/K2-Horizon-MoVA-36B-A4B liked a model 13 days ago
IFM/K2-Horizon-32BOrganizations
liked a
model 5 days ago
liked a
model 6 days ago
liked a
model 13 days ago
liked 2
models 15 days ago
reacted to RiverRider's post with 🔥 23 days ago
Post
2463
0.6B reads 27B
A 0.6B model runs in your browser tab and searches 123,287 photographs that a 27B model encoded months ago. The big model is never downloaded and never runs.
Demo: RiverRider/0.6b-reads-27b
Head: RiverRider/srt-browser-head-118k
Two models normally cannot understand each other. Each builds its own private way of representing meaning, so an image encoded by one is gibberish to the other. We fitted a small linear read-out on each frozen model, aimed at the same space. Not a fine-tune: a map outside the weights, reading a hidden state the model already computed. Now the small model's reading of your sentence compares directly against the large model's reading of a photograph.
So the 27B model's work becomes a file. It looked at the gallery once on a datacenter GPU and left about a kilobyte per photograph. Nothing of it runs when you search. The 0.6B half runs on a phone CPU through WebAssembly: no WebGPU, no API key, no server, nothing leaves the tab, and it works offline after the first load.
Privacy here is structural rather than promised, because there is no server to trust. The index is a file you hold instead of an embedding bill you pay, and it can outlive the model that made it.
Receipts. Median rank 33 of 123,287 puts the right photograph in the top 0.027%. Shuffled-pair control 0.0002. The 4 KB runtime anchor is not optional: the same head scores 0.2300 under PyTorch fp16 but 0.0154 under candle Q4, and 0.1952 once anchored.
Honest limit: word order is at chance, 0.468 on minimal pairs, n=141. "A dog chasing a boy" and "a boy chasing a dog" look alike to it. This is a bag of content, not a parser of relations.
Apache-2.0. The negatives sit in the model card next to the positives.
A 0.6B model runs in your browser tab and searches 123,287 photographs that a 27B model encoded months ago. The big model is never downloaded and never runs.
Demo: RiverRider/0.6b-reads-27b
Head: RiverRider/srt-browser-head-118k
Two models normally cannot understand each other. Each builds its own private way of representing meaning, so an image encoded by one is gibberish to the other. We fitted a small linear read-out on each frozen model, aimed at the same space. Not a fine-tune: a map outside the weights, reading a hidden state the model already computed. Now the small model's reading of your sentence compares directly against the large model's reading of a photograph.
So the 27B model's work becomes a file. It looked at the gallery once on a datacenter GPU and left about a kilobyte per photograph. Nothing of it runs when you search. The 0.6B half runs on a phone CPU through WebAssembly: no WebGPU, no API key, no server, nothing leaves the tab, and it works offline after the first load.
Privacy here is structural rather than promised, because there is no server to trust. The index is a file you hold instead of an embedding bill you pay, and it can outlive the model that made it.
Receipts. Median rank 33 of 123,287 puts the right photograph in the top 0.027%. Shuffled-pair control 0.0002. The 4 KB runtime anchor is not optional: the same head scores 0.2300 under PyTorch fp16 but 0.0154 under candle Q4, and 0.1952 once anchored.
Honest limit: word order is at chance, 0.468 on minimal pairs, n=141. "A dog chasing a boy" and "a boy chasing a dog" look alike to it. This is a bag of content, not a parser of relations.
Apache-2.0. The negatives sit in the model card next to the positives.
- 7 replies
liked 2
models 23 days ago
reacted to eaddario's post with 🔥 23 days ago
Post
2797
Experimental global target bits‑per‑weight quantization of Qwen/Qwen3.8-27B, including MTP tensors in imatrix file.
Unlike standard llama.cpp quantization that rely on fixed type heuristics (e.g., Q4_K_M), the Target BPW approach automatically optimizes per-tensor precision where it matters the most, and produces high quality models that meet a precise global size target.
Key Advantages:
- VRAM Maximization: Can generate high quality models sized exactly to fit hardware constraints (e.g., fitting the model into exactly 24GB VRAM).
- Data-Driven Precision: Quantization mix is determined by actual weight error sensitivity rather than hardcoded rules, often yielding better PPL/KLD size trade-offs.
- imatrix includes data for MTP/NextN layers.
Full benchmarks (PPL, KLD, ARC, GPQA, MMLU, etc.) and methodology in the model's card.
eaddario/Qwen3.8-27B-GGUF
Unlike standard llama.cpp quantization that rely on fixed type heuristics (e.g., Q4_K_M), the Target BPW approach automatically optimizes per-tensor precision where it matters the most, and produces high quality models that meet a precise global size target.
Key Advantages:
- VRAM Maximization: Can generate high quality models sized exactly to fit hardware constraints (e.g., fitting the model into exactly 24GB VRAM).
- Data-Driven Precision: Quantization mix is determined by actual weight error sensitivity rather than hardcoded rules, often yielding better PPL/KLD size trade-offs.
- imatrix includes data for MTP/NextN layers.
Full benchmarks (PPL, KLD, ARC, GPQA, MMLU, etc.) and methodology in the model's card.
eaddario/Qwen3.8-27B-GGUF
reacted to Banaxi-Tech's post with 😎🔥 23 days ago
Post
2707
We're releasing Overfitter 1.0.
Its a completly useless overfitted model trained on 202 epochs of SWE Bench Verified, SWE Bench Pro, Terminal Bench 2.1, DeepSWE.
It gets 100% on SWE Bench Verified, 98.6% on SWE bench Pro, 100% on Terminal Bench 2.1 and 100% on DeepSWE!
BananaMind/Overfitter-1.0
Its a completly useless overfitted model trained on 202 epochs of SWE Bench Verified, SWE Bench Pro, Terminal Bench 2.1, DeepSWE.
It gets 100% on SWE Bench Verified, 98.6% on SWE bench Pro, 100% on Terminal Bench 2.1 and 100% on DeepSWE!
BananaMind/Overfitter-1.0
- 11 replies
reacted to Nymbo's post with 🔥 about 1 month ago
Post
2210
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 2
models about 1 month ago
posted an update about 2 months ago
Post
117
deepseek-ai/DeepSeek-V4-Flash-0731 is a fantastic model, the only hint of weakness so far is in the tough instruction-following test.
For the GPU-poor, my sm89-compatible fork of vLLM is at https://github.com/the-crypt-keeper/vLLM-sm89/tree/sm89-ds4-work I am running L40S but it should work on regular L40 and 4090D as well. I have not yet tested the W2 quantization for performance loss, thats next up, so you'll need 192GB to run it.
For the GPU-poor, my sm89-compatible fork of vLLM is at https://github.com/the-crypt-keeper/vLLM-sm89/tree/sm89-ds4-work I am running L40S but it should work on regular L40 and 4090D as well. I have not yet tested the W2 quantization for performance loss, thats next up, so you'll need 192GB to run it.
liked 5
models about 2 months ago
RedHatAI/Inkling-Small-FP8-dynamic
Image-Text-to-Text • 266B • Updated • 232 • 2
antirez/deepseek-v4-gguf
Text Generation • 284B • Updated • 2.03M • 472
nazeshinjite/DeepSeek-V4-Flash-0731-ds4-GGUF
Text Generation • 20B • Updated • 2.5k • 21
bartowski/DeepSeek-V4-Flash-0731-GGUF
Text Generation • 284B • Updated • 3.4k • 40
empero-ai/Qwythos-27B-v1
Image-Text-to-Text • 28B • Updated • 2.56k • 169