fix(bench): pin TinyLlama-1.1B-Q4_0 GGUF SHA256 from first CI run#10
Merged
Conversation
marcos-mendez
added
stream-3
5 tasks
marcos-mendez
added a commit
that referenced
this pull request
May 6, 2026
…b for POPC_16A (#11) * feat(hw): inter-card connector pinout + KiCad footprint + SV port stub for POPC_16A Closes #8. Designs the 40-pin, 0.8 mm pitch dual-row board-to-board mezzanine connector that lets two POPC_16A Sails aggregate compute (multi-card parallelism mandate), even though rev-A ships single-card. Connector choice: Samtec QSE-040-01-L-D-A (or pin-compatible Hirose FX18-40P-0.8SH, JLCPCB basic part LCSC C40503). Signals only — no power between cards. Carries 4 TX diff pairs + 4 RX diff pairs + 1 forwarded clock pair + 4 sideband (RESET_N, PRSNT_N, SMB_CLK/DAT) + 13 GND + 5 reserved = 40 pins total. Width contract aligns with the MAST #14 contract verified by Spanker PR #6: INTERCARD_LANES=4, INTERCARD_LANE_WIDTH=32, INTERCARD_BUS_WIDTH=128. Artifacts: docs/hw/intercard-connector-pinout.md — full design doc, per-pin table docs/adr/0002-intercard-connector.md — decision ADR (closes issue task) docs/adr/0001-spec.md — Status amended to note ADR-002 kicad/intercard-connector/ — KiCad 8 symbol + footprint (CERN-OHL-S-2.0 via README) src/intercard_link.sv — SV port-surface stub (Apache-2.0) verif/intercard_link/ — Verilator --lint-only smoke test Smoke test passes locally with Verilator 5.048: $ bash verif/intercard_link/run_lint.sh [run_lint] PASS — intercard_link elaborates with INTERCARD_BUS_WIDTH = 128 KiCad library lives inside InnerJib7EA (not Stays) because Stays's working tree was on a stale feature branch at authoring time — collision-safe fallback documented in §5.2 of the design doc, with worked sym-lib-table snippet for Stays integration in §5.3. Out of scope (separate PRs): - PCB layout placement (Stays #10) - Controlled-impedance stackup (Stays #9) - Line coding choice (MAST ADR-014) - FPGA-side transceiver schematic capture - Hot-plug capability (rev-B) Authored by Agent 2 (FPGA Hardware). * fix(docs) * fix(adr) * fix(verif) --------- Co-authored-by: Marcos <m@pop.coop>
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.
Context
InnerJib7EA PR #9 (merged 2026-05-06 as
fb8555b) introduced the TinyLlama-1.1B reference bench harness with a deliberate bootstrap-trust gap:fetch_model.shshipped withEXPECTED_SHA256="${EXPECTED_SHA256:-PLACEHOLDER_UPDATE_AFTER_FIRST_FETCH}"so the firstbench.ymlworkflow_dispatch run would download, hash, log the value, and abort — forcing a deliberate human/Agent R pin step rather than silent first-cache-fill.What this PR does
Pins the default to the SHA256 computed by run id 25416847477 (workflow_dispatch on
fb8555b):Per first-CI-run log:
[fetch_model] computed: da3087fb.... Parameter-expansion preserved so per-invocation override viaEXPECTED_SHA256=...still works for development.Open security note (carry-over)
Bootstrap-trust point: trusting the SHA256 of whatever HuggingFace served on 2026-05-06T04:41Z. To strengthen:
Filed via the GitHub Contents API to avoid local working-tree collision (Agent 3 round 2 on Spanker, Agent 4 round 2 on Stays, Agent 1 round 2 on MAST may all be holding clones).
Authored by Agent R (Reviewer).