evo2 infer: generate CLI mode (steer from the command line)#1632
evo2 infer: generate CLI mode (steer from the command line)#1632polinabinder1 wants to merge 1 commit into
generate CLI mode (steer from the command line)#1632Conversation
Fourth CLI mode alongside serve/encode/batch: `launch_inference.sh generate --prompt ATGC... --clamp FEATURE_ID[:STRENGTH]` runs steered generation from the command line (repeat --clamp for several features). Reuses the verified engine.generate; `_parse_clamps` turns the repeatable --clamp args into feature specs. Usage docs the encode->generate steering loop. Stacked on #1622 (the engine + server), so the verified core stays frozen. test_cli.py covers the clamp parsing (CPU); the generation path itself is the engine's (GPU-verified in #1622's test_steering). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Signed-off-by: Polina Binder <pbinder@nvidia.com>
|
Important Review skippedAuto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Enterprise Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
…_sae serve` Shrink the inference PR to the engine + server + their tests. The encode/batch/generate command-line tools (cli.py) and launch_inference.sh move to the stacked CLI PR (#1632); the server stays launchable here via `python -m evo2_sae serve` (__main__.py, env-configured). fasta.py stays (shared by the extraction-side chunk_fasta.py and, via the base, the CLI). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Signed-off-by: Polina Binder <pbinder@nvidia.com>
|
Merged into the steering PR #1634, which now holds the generate CLI alongside the shared sae.steering primitive + harness, and unifies all clamping (engine/CLI/dashboard/harness) onto one implementation (sae.steering, B). |
|
Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually. Contributors can view more details about this message here. |
Summary
Adds a
generateCLI mode so you can steer SAE features from the command line — a fourthmode alongside
serve/encode/batch. Stacked on #1622 (base =pbinder/evo2-sae-serve)so the verified engine + server core stays frozen.
What's here (diff vs #1622)
cli.py:generatesubparser + handler (reusesengine.generate);_parse_clampsturnsrepeatable
--clamp FEATURE_ID[:STRENGTH]args into feature specs.launch_inference.sh: usage now documentsgenerate+ the encode→generate steering loop.tests/test_cli.py: CPU coverage of_parse_clamps(3 tests, pass).Tests / verification
_parse_clamps— 3 CPU tests pass (id:strength, default strength, empty).generate, GPU-verified in evo2 SAE recipe: live inference engine + steering server + CLI #1622 (test_steering).Merge #1622 first; GitHub auto-retargets this to main afterward.
🤖 Generated with Claude Code