Solana Clawd LoRA
LoRA adapter trained from Qwen/Qwen2.5-1.5B-Instruct on data/model_kit/clawd_future_drill_processed.
Hub model ID: solanaclawd/clawd-solana-masterpiece-qwen15-lora
Training
- Train rows: 583
- Eval rows: 32
- Max sequence length: 768
- LoRA rank/alpha: r=16, alpha=32
- LoRA target modules:
q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj - Output directory:
outputs/clawd-solana-masterpiece-qwen15-lora-mac-v3 - W&B run: https://wandb.ai/clawdsolana-clawd/clawd-ai-training/runs/kvol8jip
Metrics
| Metric | Value |
|---|---|
epoch |
0.329331 |
total_flos |
4.4773e+14 |
train_loss |
0.285656 |
train_runtime |
74.1024 |
train_samples_per_second |
2.591 |
train_steps_per_second |
2.591 |
Intended Use
This adapter is intended for Solana-native Clawd agents that need project-local context around core-ai, Helius integrations, Clawd Code, Clawd Grok, MCP server conventions, agent skills, and the existing Solana/DeFi/ZK instruction corpus.
Loading
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
base_model = "Qwen/Qwen2.5-1.5B-Instruct"
adapter_id = "solanaclawd/clawd-solana-masterpiece-qwen15-lora"
tokenizer = AutoTokenizer.from_pretrained(base_model, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(base_model, device_map="auto", trust_remote_code=True)
model = PeftModel.from_pretrained(model, adapter_id)
Safety
The dataset builder runs in public-safe mode by default and excludes common secret filenames, private key/token patterns, binary artifacts, dependency folders, lockfiles, and high-risk security records that are not suitable for public dataset release.
This adapter is a research/developer artifact. Live trading or wallet actions must remain behind separate execution clients, simulation, explicit operator approval, and pre-trade risk gates.
- Downloads last month
- 13