marcuspaico
0bdae43455
feat: Gemini Flash judge harness + agreement subsets
...
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com >
2026-08-17 21:15:24 -07:00
marcuspaico
e92782639a
fix: align collect_frontier token budget with driver (8k)
...
kimi-k3 spends heavily on hidden reasoning tokens that draw from the same
budget as the visible answer. 3000-token budget resulted in empty responses
on hard prompts. Increased to 8000 to accommodate reasoning overhead while
ensuring room for substantive visible answers.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com >
2026-08-17 20:54:34 -07:00
marcuspaico
b7d0f38ce8
fix: raise kimi-k3 token budget to 8k (reasoning tokens exhausted 3k on hard prompts)
...
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com >
2026-08-17 20:51:32 -07:00
marcuspaico
75abea6431
feat: CompassionBench bank + 4-system response collection
2026-08-17 20:44:11 -07:00
marcuspaico
7a30e06ab0
feat: QLoRA fine-tune v1 on M5 + fused model (M2)
2026-08-17 13:43:32 -07:00
marcuspaico
d55d9a648c
fix: complete template coverage in training data + disclose model alias
2026-08-15 15:50:27 -07:00
marcuspaico
cc86a97d10
docs: record verified tilde-alias model ID in constraints
...
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com >
2026-08-15 14:25:27 -07:00
marcuspaico
7b97992b1b
feat: synthetic instruction generation via OpenRouter deepseek-v4-flash
2026-08-15 14:18:37 -07:00
marcuspaico
1b2d84094d
chore: gitignore .openrouter_key
...
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com >
2026-08-14 17:45:53 -07:00
marcuspaico
e282fe5631
docs: fix remaining Claude reference in spec diagram
...
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com >
2026-08-14 17:40:13 -07:00
marcuspaico
6d6fa8f163
docs: switch API layer to OpenRouter (deepseek-v4-flash gen, gemini-flash judge, kimi-k3 reference)
...
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com >
2026-08-14 17:40:01 -07:00
marcuspaico
37caa87b49
feat: RAG answers with sutta citations (M1)
...
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com >
2026-08-14 17:26:26 -07:00
marcuspaico
c14b314107
test: add integration test for build_index/search contract (uid/title/chunk/score)
...
Adds test_build_index_and_search_roundtrip to verify:
- Index building and search round-trip with fixture JSONL
- Result keys exactly match {uid,title,chunk,score} contract
- Anger sutta ranks first for anger-related query
Closes plan-gap: regression coverage for Task 4 dependencies.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com >
2026-08-14 17:22:13 -07:00
marcuspaico
691d2d94f5
feat: chunked bge embeddings + lancedb sutta index
...
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com >
2026-08-14 17:18:42 -07:00
marcuspaico
221b1f5097
fix: collection field extraction and add regression tests for sort-key and nested layout
2026-08-14 17:11:47 -07:00
marcuspaico
f456877b58
feat: bilara corpus ingestion to suttas.jsonl
2026-08-14 17:06:26 -07:00
marcuspaico
76d1858ffe
chore: scaffold buddhagpt, verify Qwen2.5-7B-4bit runs on M5 (PAI-87)
2026-08-14 17:01:33 -07:00
marcuspaico
4a5f2026c4
docs: BuddhaGPT design spec + implementation plan (PAI-87)
...
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com >
2026-08-14 16:41:19 -07:00
