serve: engine default still said glm after the rename — direct invocation broken on a clean checkout - #526
Merged
Merged
Conversation
python3 openai_server.py --model <dir> looked for a binary named 'glm' next to itself. Since #391 the build produces 'colibri', so direct invocation was broken on any clean checkout -- it only appeared to work in trees that still had a stale glm from an older build. Spotted by @RDouglasSharp while working on #488. Resolve the engine by probing colibri / colibri.exe first and falling back to glm / glm.exe, the same order the coli launcher uses, so old trees keep starting. Verified by removing the stale glm and running the gateway: it resolves to colibri and the engine launches. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
pull Bot
pushed a commit
to danielabelski/colibri
that referenced
this pull request
Jul 22, 2026
v1.1.0 trips Microsoft Defender on Windows, and the cause was ours: 107 KB of zeros in .data (JustVugg#527/JustVugg#532). This gets the fix out rather than leaving every new Windows download to hit the same alert. Also carries the Anthropic /v1/messages endpoint (JustVugg#343), the gateway engine default fix (JustVugg#526), release checksums (JustVugg#530), and the Get-started docs rewrite (JustVugg#521). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This was referenced Jul 22, 2026
gohlerdev
pushed a commit
to gohlerdev/BetterColibri
that referenced
this pull request
Jul 28, 2026
The tiny-model efficiency/regression suite silently skipped on every tree since the JustVugg#391 rename: it looked for glm.exe (test_inefficiency ENGINE, efficiency.run_engine default, test_cuda_env GLM) while the build produces colibri — the exact bug class JustVugg#526 fixed in the gateway. All three now use the gateway's resolution order (colibri, colibri.exe, glm, glm.exe) with actionable skip messages. Executing the suite for the first time exposed a flaky assertion: test_cpu_vs_cpu_determinism bounded run-to-run tok/s drift at 25%, but a tiny-model decode is ~20 ms of wall clock — scheduler noise, not an engine property (measured >2x variance, ~1 failure in 3 runs). The determinism contract keeps its exact greedy hit-rate equality (the part that catches real non-determinism) and timing gets a sanity bound (positive, finite) instead of a cross-run tolerance. Validation: with the glm_tiny fixture present the suite now RUNS (5 tests executed, was 8/8 skipped) and passes 5/5 consecutive runs; full test-python 143 OK. Teacher-forcing oracle validated alongside: TF=1 vs transformers reference = 32/32 at f32.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Noticed by @RDouglasSharp in passing while debugging #488, and it reproduces:
python3 openai_server.py --model <dir>looks for a binary namedglmbeside itself, but since #391 the build producescolibri. It only seemed to work for those of us with a staleglmleft over from an older build — on a clean checkout the gateway cannot find its engine.coli servewas unaffected (the launcher resolves the engine itself and passes--engine), which is why CI never caught it.Fix: resolve by probing
colibri/colibri.exefirst, falling back toglm/glm.exe— the same order thecolilauncher uses, so old trees keep starting. Verified by moving the staleglmout of the way: resolves tocolibri, engine launches. Python suite green (149).Thanks @RDouglasSharp — reporting a second bug you weren't even looking for is appreciated.
🤖 Generated with Claude Code