Repository navigation
Conversation
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Add opt-in, JSON-compatible scalar calibration snapshots across identical model owners:
Lazy per-agent defaults and missing-layer calibration remain unchanged. Records retain no model, tensors, KV cache, or GPU state. No filesystem service is added.
Evidence
Full external-process wall medians, three fresh counterbalanced process pairs per workload:
A = pinned Qwen3-0.6B; B = pinned Qwen3-1.7B. RTX 4090, unquantized BF16, Torch 2.8.0, Transformers 4.57.1, SDPA, four threads. Each owner runs two 16-token public turns including NLL. Timings include initial calibration misses, key/profile/JSON/import/export work, model loads, first-use warmup, release, and interpreter startup/teardown. Filesystem warm; downloads, cold-disk reads, and SSH excluded. This measures startup/session throughput, not decode acceleration; no confidence interval or general guarantee.
CUDA comparisons exactly match calibration, scaled activations, first logits, token IDs, NLL, and public state across 45 controls. CPU tests cover cold hits, defaults, atomic malformed-import recovery, reset, adapter keys, tracebacks, and owner lifetime. Existing selected-model GPU tests pass; the no-BOS case skips.
Design Choices
ValueErrorand leave all norms empty. Reusable records require canonical zero-based layers; existing negative indexing remains default-only.