Downloads · 30 days
12
80% of all-time downloads
wallfacers/engram-planner-lora
engram-planner-lora is a text generation model from wallfacers. Use it when you need the model to write or continue text. It is set up for peft.
本适配器是 engram 的本地 Evidence Planner:给定问题 + 候选证据列表,输出 022 Evidence Compiler 合同的 proposal JSON(need + KEEP/gap actions)。训练数据 = 023 spec r5 冻结版本。spec:specs/023-local-trained-evidence-compiler/。
Downloads · 30 days
12
80% of all-time downloads
All-time downloads
15
Public
Repo size
51.8 MB
Likes
0
Public
Click a slice to open those files.
.safetensors40.4 MB · 78%
From the Hugging Face model README
本适配器是 engram 的本地 Evidence Planner:给定问题 + 候选证据列表,输出 022
Evidence Compiler 合同的 proposal JSON(need + KEEP/gap actions)。训练数据 = 023 spec
r5 冻结版本。spec:specs/023-local-trained-evidence-compiler/。
Qwen/Qwen2.5-7B-Instruct(Apache-2.0)r=16, lora_alpha=32, lora_dropout=0.05, 目标模块 q/k/v/o_proj023-b20260803-r5,T011 门 PASS(199/200 = 99.5%,Wilson [97.2%, 99.9%])# vLLM LoRA 模式(OpenAI-compatible sidecar)
serve.sh lora --model Qwen2.5-7B-Instruct --adapter models/planner-lora
# 或合并底模(post-merge snapshot)
推理 prompt 模板必须与训练一致(cmd/locomo-bench/local_planner.go 的
plannerSystemPrompt + renderPlannerPrompt;train_lora.py 的 SYSTEM_PROMPT 镜像它)。