Kral d5e43e1a83 Trajectory record (messages, raw tool results, metadata; reasoning apart) and trajectory runner ...

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

2026-10-05 12:03:54 +02:00

docs

Step 400 check: Mac -13.4 % vs GPU -35.0 %; next GPU run bf16 mixed with stage 2; findings in STATE and roadmap

2026-10-05 11:46:21 +02:00

harness

Trajectory record (messages, raw tool results, metadata; reasoning apart) and trajectory runner

2026-10-05 12:03:54 +02:00

tasks

Runner: G2 finds the FUNCTION statement after local classes; budget floor applied to all tasks

2026-10-03 18:25:30 +02:00

tasks_gen/eval

G0181 failure analysis (int8 in SALV), W7B results

2026-10-03 19:16:56 +02:00

train

Step 400 check: Mac -13.4 % vs GPU -35.0 %; next GPU run bf16 mixed with stage 2; findings in STATE and roadmap

2026-10-05 11:46:21 +02:00

.env.example

Harness: runner, generator, tasks T01 T13 T14 T15, CLAUDE.md

2026-10-02 21:42:51 +02:00

.gitignore

Stage 1 step 1: prepare.py, real corpus numbers (SAP-samples/abap-cheat-sheets)

2026-10-03 21:58:04 +02:00

CLAUDE.md

Stage 1 on HF Jobs: Unsloth job script, PEFT to MLX converter, base valid loss 0.849, Qwen base model docs

2026-10-04 18:51:40 +02:00

package-lock.json

Harness: runner, generator, tasks T01 T13 T14 T15, CLAUDE.md

2026-10-02 21:42:51 +02:00

package.json

Harness: runner, generator, tasks T01 T13 T14 T15, CLAUDE.md

2026-10-02 21:42:51 +02:00

Description

ABAP LLM Training

2.4 MiB

Languages

ABAP 90.6%

Python 6.3%

ABAP CDS 3%

Shell 0.1%