Qwen3.8
Collection
The best • 10 items • Updated
🔄 In a Training Loop
Zxbc
zxbc2023
AI & ML interests
I make things less shitty.
Recent Activity
new activity about 1 hour ago
ornith-ai/Ornith-1.5-35B-A3B:Best local model that runs on commodity hardware updated a collection about 6 hours ago
Qwen3.8 updated a collection about 6 hours ago
Qwen3.8Organizations
Best local model that runs on commodity hardware
😔 1
10
#23 opened 2 days ago
by 
dwidlee
updated a
collection about 6 hours ago
reacted to dejanseo's post with 😔 about 16 hours ago
Post
2923
Ox Alpha is GLM
https://dejan.ai/blog/ox-alpha/
A parameter-free k-nearest-neighbour classifier over Normalized Compression Distance (Lee et al., MobiSys ’24, Eq. 1, built on Jiang et al.'s gzip-based text classifier). NCD compares two texts by how well they compress together. C(s) is the gzip-compressed length of s. Text sharing an author's patterns compresses better together than text from a different author, so the method needs no model weights and no embeddings.
The reference corpus covers 60 prompts (essays, code, emails, dialogue, poetry) answered by five known models: GPT-5.5, Claude Opus 5, Gemini 3.7 Flash, Gemini 3.1 Pro Preview, and GLM-5.3, for 293 reference texts. ox-alpha answered the first 13 of those prompts, plus one additional novel prompt never given to the reference models beforehand, for 14 queries in total. Each query was classified against the reference corpus independently, with a k-nearest-neighbour vote (k=5):
Model ox-alpha samples matched
GLM-5.3 7 / 14
Claude Opus 5 3 / 14
Gemini 3.7 Flash 2 / 14
GPT-5.5 1 / 14
Gemini 3.1 Pro Preview 1 / 14
GLM-5.3 wins at every k tested: 7/14 at k=3, 7/14 at k=5, 6/14 at k=7, 7/14 at k=9. Claude Opus 5 is the consistent second place.
https://dejan.ai/blog/ox-alpha/
A parameter-free k-nearest-neighbour classifier over Normalized Compression Distance (Lee et al., MobiSys ’24, Eq. 1, built on Jiang et al.'s gzip-based text classifier). NCD compares two texts by how well they compress together. C(s) is the gzip-compressed length of s. Text sharing an author's patterns compresses better together than text from a different author, so the method needs no model weights and no embeddings.
The reference corpus covers 60 prompts (essays, code, emails, dialogue, poetry) answered by five known models: GPT-5.5, Claude Opus 5, Gemini 3.7 Flash, Gemini 3.1 Pro Preview, and GLM-5.3, for 293 reference texts. ox-alpha answered the first 13 of those prompts, plus one additional novel prompt never given to the reference models beforehand, for 14 queries in total. Each query was classified against the reference corpus independently, with a k-nearest-neighbour vote (k=5):
Model ox-alpha samples matched
GLM-5.3 7 / 14
Claude Opus 5 3 / 14
Gemini 3.7 Flash 2 / 14
GPT-5.5 1 / 14
Gemini 3.1 Pro Preview 1 / 14
GLM-5.3 wins at every k tested: 7/14 at k=3, 7/14 at k=5, 6/14 at k=7, 7/14 at k=9. Claude Opus 5 is the consistent second place.
- 6 replies
Investigating the Reasoning vs. Knowledge Trade-off in Ornith-1.5
👍 1
9
#26 opened 1 day ago
by 
AnbaiMei
Great work. Bad advertisement.
👍 5
2
#17 opened 4 days ago
by 
Izenflamm
No speculative decoding, No QAT
8
#21 opened 2 days ago
by 
engrtipusultan
liked a
model 1 day ago
updated a
Space 1 day ago
liked 2
models 2 days ago
published a
Space 2 days ago
in my local heavy agentic work your model is way better
❤️ 3
3
#2 opened 3 days ago
by 
ntoytoy
Code acceptance rate and token speed drops on write tool calls in pi coding agent
👍➕ 3
6
#3 opened 6 days ago
by 
sandeshrajx
updated a
Space 4 days ago
published a
Space 4 days ago
Seems to perform worse than 1.0
👍 3
1
#13 opened 4 days ago
by 
russellcg
mtp.* tensors look like random init, not trained weights
🧠👍 8
14
#10 opened 5 days ago
by 
EryriLabs
🚩 Report: Illegal or restricted content
🧠🤯 3
8
#3 opened 5 days ago
by 
seadownfallpioneer10