Gabriel Devenyi
gdevenyi
AI & ML interests
None yet
Recent Activity
new activity 6 days ago
Intel/Qwen3.8-Flash-Next-W4A16-AutoRound:model-00002-of-00017.safetensors is missing liked a model 10 days ago
RadixArk/Qwen3.8-27B-NVFP4-BF16-LMHeadOrganizations
model-00002-of-00017.safetensors is missing
1
#1 opened 6 days ago
by
gdevenyi
Add a `terse` template kwarg to opt out of the terseness block
❤️ 1
2
#5 opened 13 days ago
by
gdevenyi
Add a `_default_reasoning_effort` knob and remove the dead `_initial_effort` line
1
#91 opened 13 days ago
by
gdevenyi
Make reasoning effort steering more intuitive in the Jinja template
3
#3 opened 15 days ago
by
extrabigmehdi
v22.3 release notes
👍🔥 8
8
#85 opened 14 days ago
by
froggeric
README: add vLLM setup and correct the top-level reasoning_effort claim
❤️ 1
1
#4 opened 13 days ago
by
gdevenyi
README: qualify the official enable_thinking claim, document sticky inline tags and the string-argument fallback
1
#90 opened 13 days ago
by
gdevenyi
README: correct vLLM setup and document which thinking-off channels vLLM's parser tracks
1
#89 opened 13 days ago
by
gdevenyi
Read the assistant `reasoning` field (vLLM / Responses API name) when rendering history
1
#88 opened 13 days ago
by
gdevenyi
Use a single newline between consecutive tool calls (matches official Qwen templates)
1
#87 opened 13 days ago
by
gdevenyi
Speed when CPU-only?
8
#51 opened 22 days ago
by
TriAxp
Full review
#20 opened 27 days ago
by
gdevenyi
Plans for w4a16 4-bit model?
1
#1 opened about 1 month ago
by
gdevenyi
A comprehensive review
👍 1
#48 opened about 1 month ago
by
gdevenyi
FP8 quant?
3
#8 opened about 2 months ago
by
gdevenyi
FP8 variant possible?
2
#1 opened 3 months ago
by
maglat
Repo doesn't build
4
#2 opened 4 months ago
by
gdevenyi
Please rename files
#1 opened 4 months ago
by
gdevenyi
mmproj supported?
2
#1 opened 4 months ago
by
gdevenyi