Clone-to-run packaging + finish plan for the SL2610 FunctionGemma demo#9
Open
michalharakal wants to merge 4 commits into
Open
Clone-to-run packaging + finish plan for the SL2610 FunctionGemma demo#9michalharakal wants to merge 4 commits into
michalharakal wants to merge 4 commits into
Conversation
Bump the demo's transformers BOM pin to the released 0.35.0, which re-exports sk.ainet:skainet-bom so the engine sk.ainet.core:* align to 0.35.0 too. Verified: jvmCompileClasspath resolves every sk.ainet.transformers:* and sk.ainet.core:* to 0.35.0 from Maven Central (the FunctionGemma work + argMax op). The local vendor plugin sk.ainet.vendors:synaptics-torq:0.1.0 stays composite-only (unpublished). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
In-flight KV-cache decode program (board-UNVERIFIED drafts), frozen before the clone-to-run packaging work so the baseline is reproducible: - GemmaKvDecoder wiring in Pipeline.kt (GEMMA_KV=1) + MoonshineKvDecoder + MoonshineRunner (MOONSHINE_KV=1) - GEMMA-KV-BOARD-LOOP.md, PERF-LOGBOOK.md The matching GEMMA_KV/GEMMA_QUANT build logic in scripts/compile-gemma.sh is interleaved with packaging edits and lands whole in the next commit. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…n docs Make the demo clone-to-run and extensible without path/IP surgery: - demo.env(.example): single config surface (BOARD, TORQ_PKG, GEMMA_GGUF, board dirs). Every scripts/*.sh sources it; inline env still wins; no hardcoded personal paths. - bootstrap.sh: one-time setup (demo.env, locate Torq toolchain, model checks). - README: Quick-start (clone -> bootstrap -> deployBoard) + layout. - docs/TOOLCHAIN-PIN.md: g165e12a-vs-v2.0.0 currency decision + canary gate. - docs/FINETUNING.md + examples/custom-command/: teach FunctionGemma new commands (Octopus-v2 dataset -> LoRA -> GGUF -> compile-gemma.sh -> deploy). - docs/BOARD-RUNBOOK.md: board-dependent steps (self-compiled default, KV verify) + repo extraction. - .gitignore: demo.env, .toolchain/, weights/. scripts/compile-gemma.sh also carries the in-flight GEMMA_KV/GEMMA_QUANT graph build (interleaved with the config-externalization edits); committed whole here. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The finish plan (P0-P5) was a loose file at the untracked workspace root. Move it into the demo's docs/ so it's version-controlled alongside the packaging it describes; reword its path-orientation note for the new home and repoint the README links to docs/FINISH-PLAN.md. The workspace-root PLAN.md (still loose) now links to the new path. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
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.
Makes the demo clone → connect board → run → finetune on your own commands, and lands the
forward-looking finish plan. Also carries the in-flight KV-cache decode drafts (frozen) and the
0.35.0 dependency bump.
What's in here (4 commits)
skainetTransformers0.34.0 → 0.35.0 (shipsmoonshineDecoder()+ FunctionGemma self-compile).MoonshineKvDecoder+GemmaKvDecoderwiring (GEMMA_KV=1/MOONSHINE_KV=1),docs/GEMMA-KV-BOARD-LOOP.md,docs/PERF-LOGBOOK.md. Board-UNVERIFIED drafts (opt-in behind env flags; default paths unchanged).demo.env(.example): single config surface (BOARD,TORQ_PKG,GEMMA_GGUF, board dirs). Everyscripts/*.shsources it; inline env wins; no hardcoded personal paths / board IP.bootstrap.sh+ README Quick-start (clone → bootstrap → deployBoard).docs/TOOLCHAIN-PIN.md(g165e12a-vs-v2.0.0 currency decision + canary gate),docs/FINETUNING.md+examples/custom-command/(teach FunctionGemma new commands),docs/BOARD-RUNBOOK.md..gitignore:demo.env,.toolchain/,weights/.docs/.Notes for review
scripts/compile-gemma.shcarries both the config-externalization edits and the in-flightGEMMA_KV/GEMMA_QUANTgraph build (interleaved in the same hunks — committed whole).feature/google-io-demo(the current integration branch; the prior baseencoder-npu-rope-seqfrom PR feat(gemma): self-compile FunctionGemma — compile-gemma.sh, retire Python #8 was deleted after merge). Retarget tomainif a direct-to-main land is preferred.docs/BOARD-RUNBOOK.md.🤖 Generated with Claude Code