Skip to content
Draft
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
25 changes: 25 additions & 0 deletions notes/pr6_update_summary.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,25 @@
# PR #6 Update Summary

## What this PR adds

- an external-handoff intake plan for the selected Runpod path
- a resume contract that preserves the unchanged TPI-004 execution commands
- a handoff decision surface for whether execution can resume immediately
- an explicit absent-state record when no new provider handoff arrives

## What still remains

- concrete external provider handoff data
- verified landing in `/workspace/parameter-golf`
- actual resume of the TPI-004 evidence pass

## First required external handoff

The next turn should begin from the exact provider-supplied attach / SSH route and immediately try to verify landing in `/workspace/parameter-golf`.

## This turn's outcome

- no new external handoff data was present
- no attach or landing verification was attempted
- no local provider discovery was reopened
- TPI-004 remains paused behind the same external handoff requirement
59 changes: 59 additions & 0 deletions notes/tpi_007_handoff_decision.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,59 @@
# TPI-007 Handoff Decision

## Status

absent

## Objective

Record whether concrete external provider handoff data has been received and whether the unchanged TPI-004 evidence pass can now resume.

## Required decision fields

- handoff package received or not: not received
- landing path verified or not: not verified
- TPI-004 resume ready or not: not ready

## Classification

- `accepted`
- `partial`
- `absent`

## External handoff package received

- no

## Concrete handoff data received

- none in this turn

## Landing verification

- attempted: no
- `/workspace/parameter-golf` verified: no

## Resume readiness

- `resume_ready: false`

## Classification result

- `absent`

## Interpretation

- No new provider-supplied attach command, SSH tuple, endpoint, username, port, or pod id was provided in this turn.
- TPI-007 therefore stops at intake gating rather than reopening local discovery.
- TPI-004 remains suspended behind the same external handoff requirement.

## Minimum package required for acceptance

- exact provider-supplied attach command, or
- exact SSH tuple with host, username, and port, plus
- pod identifier or display name, and
- expected landing path confirmation for `/workspace/parameter-golf`

## Resume condition

TPI-004 should resume immediately once the handoff package is concrete enough to attach and verify `/workspace/parameter-golf`.
42 changes: 42 additions & 0 deletions notes/tpi_007_handoff_intake.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,42 @@
# TPI-007 Handoff Intake

## Objective

Accept concrete external provider handoff data and use it immediately to resume the suspended TPI-004 evidence pass.

## Public-facing name

`MonkeyModel_EvalFirst_HandoffIntake`

## Required external handoff fields

- pod identifier or display name
- host / endpoint
- exact attach or SSH command
- username
- port (if needed)
- expected landing path

## Acceptance rule

Only concrete provider-supplied handoff data counts as progress in this loop. Repeating local discovery without new external data does not.

## Intake result for this turn

- external handoff package received: no
- concrete provider-supplied attach command received: no
- concrete SSH tuple received: no
- landing-path verification attempted: no

## Interpretation

- No new external handoff data was present in this turn's user request.
- No new concrete attach route was embedded in the current branch notes.
- Per TPI-007 scope, no new local discovery was started.

## Resume target

If the handoff is sufficient, the next immediate action is:
1. attach
2. verify landing in `/workspace/parameter-golf`
3. resume the unchanged TPI-004 baseline/candidate evidence pass
70 changes: 70 additions & 0 deletions notes/tpi_007_resume_contract.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,70 @@
# TPI-007 Resume Contract

## Objective

Describe the minimum concrete external handoff needed to resume the unchanged TPI-004 evidence pass.

## Required handoff package

- pod identifier or display name
- host / endpoint
- exact attach or SSH command
- username
- port (if needed)
- expected landing path

## First commands after successful handoff

```bash
pwd
ls /workspace
cd /workspace/parameter-golf
git rev-parse --abbrev-ref HEAD
python3 -c "import torch, datasets, sentencepiece; print('deps-ok')"
nvidia-smi
```

## Resume commands after verification

### Baseline
```bash
RUN_ID=tpi004_baseline_stride1024 \
DATA_PATH=/workspace/parameter-golf/data/datasets/fineweb10B_sp1024/ \
TOKENIZER_PATH=/workspace/parameter-golf/data/tokenizers/fineweb_1024_bpe.model \
VOCAB_SIZE=1024 \
TRAIN_SEQ_LEN=1024 \
EVAL_STRIDE=1024 \
MAX_WALLCLOCK_SECONDS=600 \
TRAIN_LOG_EVERY=50 \
VAL_LOSS_EVERY=200 \
torchrun --standalone --nproc_per_node=1 train_gpt.py
```

### Candidate
```bash
RUN_ID=tpi004_candidate_stride128 \
DATA_PATH=/workspace/parameter-golf/data/datasets/fineweb10B_sp1024/ \
TOKENIZER_PATH=/workspace/parameter-golf/data/tokenizers/fineweb_1024_bpe.model \
VOCAB_SIZE=1024 \
TRAIN_SEQ_LEN=1024 \
EVAL_STRIDE=128 \
MAX_WALLCLOCK_SECONDS=600 \
TRAIN_LOG_EVERY=50 \
VAL_LOSS_EVERY=200 \
torchrun --standalone --nproc_per_node=1 train_gpt.py
```

## Gate

Do not reopen environment-selection or provider-discovery work if the concrete handoff package is provided and valid. Resume execution immediately.

## Current turn status

- handoff package present: no
- attach may be attempted this turn: no
- landing verification may be attempted this turn: no
- TPI-004 unchanged resume may start this turn: no

## Absence rule

If the turn contains no new provider-supplied handoff data, stop at the decision note and preserve the unchanged TPI-004 execution contract for the next intake turn.
10 changes: 10 additions & 0 deletions runs/TPI-004/run_notes.md
Original file line number Diff line number Diff line change
Expand Up @@ -110,3 +110,13 @@ blocked before pod attachment
- no username or port found
- implication:
- TPI-004 remains blocked on external provider handoff, not on the eval-first monkey-model branch itself

## TPI-007 external handoff gate update

- branch checked for intake gate: `exp/eval-first-007`
- current local commit during intake gating: `fd91b709ee29c7126a76ddc7a8df594eb769c6e7`
- new external handoff package in this turn: no
- attach attempted: no
- landing verification attempted: no
- implication:
- TPI-004 remains paused unchanged until one concrete provider-supplied attach route is handed in