初始化项目,由ModelHub XC社区提供模型
Model: Delentia/delentia-slm-jitna-v0.4 Source: Original Platform
This commit is contained in:
34
training_config/lora_config.yaml
Normal file
34
training_config/lora_config.yaml
Normal file
@@ -0,0 +1,34 @@
|
||||
# LoRA adapter configuration — standalone reference for inference/export
|
||||
# Matches the training config in slm_jitna_v0.1.yaml
|
||||
|
||||
peft_type: "LORA"
|
||||
task_type: "CAUSAL_LM"
|
||||
|
||||
# Adapter dimensions
|
||||
r: 16
|
||||
lora_alpha: 32
|
||||
lora_dropout: 0.05
|
||||
bias: "none"
|
||||
use_rslora: true
|
||||
|
||||
# Target modules (all projection layers)
|
||||
target_modules:
|
||||
- q_proj
|
||||
- k_proj
|
||||
- v_proj
|
||||
- o_proj
|
||||
- gate_proj
|
||||
- up_proj
|
||||
- down_proj
|
||||
|
||||
# Inference settings
|
||||
inference_mode: false # set true for inference-only loading
|
||||
|
||||
# Merge settings (for GGUF export)
|
||||
merge_before_export: true
|
||||
export_format: "Q4_K_M" # 4-bit quantized GGUF for Ollama
|
||||
|
||||
# Paths (relative to repo root)
|
||||
adapter_path: "models/adapters/jitna_v0.1"
|
||||
merged_path: "models/merged/jitna_v0.1"
|
||||
gguf_path: "models/gguf/delentia-jitna-v0.1-Q4_K_M.gguf"
|
||||
Reference in New Issue
Block a user