81 lines
2.5 KiB
Text
81 lines
2.5 KiB
Text
# Pere-run rerun2 pipeline — BASE ARM (glm-5 editor). Byte-faithful copy of the prod stack
|
|
# backend/configs/pipeline-c1.yaml @HEAD (D38.5), with ONLY these contract-justified deviations
|
|
# (config lives outside backend/, so relative paths would resolve wrong):
|
|
# 1. prompt paths ABSOLUTE (../prompts would resolve to /home/ubuntu/books/... — miss).
|
|
# Content = the same backend/prompts files → prompt_sha256 in the snapshot is unchanged.
|
|
# 2. escalation.budget_usd = 3.0 (>0): pipeline-c1 holds 0 for green CI; the acceptance run MUST
|
|
# be >0 or the single-hop draft echo-escalation (deepseek-v4-flash -> deepseek-v4-pro, D18/D22.11)
|
|
# never fires. budget_usd is NOT snapshot-folded (pure money-gate); ceilings live in book.yaml.
|
|
# 3. mining.contrast_path added (absolute jieba dict) so the W1.5 bank-mining stop is armed.
|
|
# contrast_path is NOT snapshot-folded. Mining also needs langpack_root set in book.yaml.
|
|
# 4. waves.workers 4 = transport parallelism only (NOT snapshot-folded); rate-guard in models.yaml
|
|
# still caps mistral concurrency in that arm.
|
|
# Everything else (models, prompt_version, temperature, reasoning, gates, sanitizer, segmentation,
|
|
# escalation chains) is identical to pipeline-c1 @HEAD. Editing prompt/version = resnapshot = re-pay.
|
|
core: C1
|
|
version: 1
|
|
|
|
defaults:
|
|
max_output_ratio: 2.2
|
|
min_max_tokens: 2048
|
|
|
|
context:
|
|
glossary_injection: selective
|
|
glossary_token_budget: 800
|
|
cache_ttl: "5m"
|
|
|
|
segmentation:
|
|
draft_budget_out: 1797
|
|
edit_ceiling_out: 3200
|
|
fertility:
|
|
cjk: 1.1978
|
|
other: 0.3852
|
|
|
|
waves:
|
|
workers: 4
|
|
|
|
retries:
|
|
regenerate_before_escalate: 1
|
|
|
|
stages:
|
|
- name: draft
|
|
role: translator
|
|
model: deepseek-v4-flash
|
|
prompts:
|
|
zh-ru: /home/ubuntu/projects/textmachine/backend/prompts/translator.md
|
|
prompt_version: v1-reflow
|
|
temperature: 0.3
|
|
reasoning: "off"
|
|
escalate_to: deepseek-v4-pro
|
|
- name: edit
|
|
role: editor
|
|
model: glm-5
|
|
prompts:
|
|
zh-ru: /home/ubuntu/projects/textmachine/backend/prompts/editor.md
|
|
prompt_version: v3-discourse-reflow
|
|
temperature: 0.4
|
|
reasoning: "off"
|
|
|
|
gates:
|
|
coverage:
|
|
enabled: false
|
|
len_ratio_bounds:
|
|
zh-ru: [2.2, 4.2]
|
|
ja-ru: [1.4, 2.6]
|
|
en-ru: [0.70, 1.4]
|
|
sent_cov_min: 0.75
|
|
min_chunk_chars: 500
|
|
sanitizer:
|
|
enabled: true
|
|
|
|
escalation:
|
|
chains:
|
|
default: [deepseek-v4-pro, glm-5.1, gemini-3.1-pro-preview]
|
|
adult: [grok-4.3]
|
|
budget_usd: 3.0
|
|
|
|
fanout:
|
|
candidates: 1
|
|
|
|
mining:
|
|
contrast_path: /home/ubuntu/projects/textmachine/eval/exp16/data/jieba_dict_general_zh.txt
|