kosiasuzu/agenticml-llama3.1-8b-lora-adapter

kosiasuzu/telos-llama3.1-8b-lora-adapter

LoRA adapter only for Telos-format agent fine-tuning on Llama 3.1 8B. Load on top of `kosiasuzu/telos-agent-llama-3.1-8b-init`.

For single-file inference without PEFT, use the merged checkpoint: `kosiasuzu/telos-llama3.1-8b-lora-merged`.

Model Details

Model Description

• —Model ID: kosiasuzu/telos-llama3.1-8b-lora-adapter
• —Developed by: Telos project / kosiasuzu
• —Model type: PEFT LoRA adapter (CAUSAL_LM)
• —Base model: kosiasuzu/telos-agent-llama-3.1-8b-init (required at load time)
• —License: Llama 3.1 Community License (via base model)

Training: same run as the Telos merged model — frames column on kosiasuzu/telos-agent-trajectory-dataset, TelosTokenizer, assistant/model-block label masking only.

Model Sources

• —Telos repository: https://github.com/kosiasuzu/talos
• —Merged weights: `kosiasuzu/telos-llama3.1-8b-lora-merged`
• —Dataset: `kosiasuzu/telos-agent-trajectory-dataset`

Uses

Direct Use

• —Load with PEFT on the init base for Telos trajectory continuation or further fine-tuning.
• —Swap or stack adapters only if you manage compatibility (this adapter is trained for one base revision).

Downstream Use

• —Merge locally: model.merge_and_unload() (as in telos training push path).
• —Eval: telos eval-format-validity --format telos --adapter-mode peft --model <init> --adapter-id kosiasuzu/telos-llama3.1-8b-lora-adapter

Out-of-Scope Use

• —Will not work if loaded on meta-llama/Llama-3.1-8B or ChatML init — base must be telos-agent-llama-3.1-8b-init.
• —Not intended for ChatML apply_chat_template workflows.

How to Get Started with the Model

import torch
from transformers import AutoModelForCausalLM
from peft import PeftModel

base_id = "kosiasuzu/telos-agent-llama-3.1-8b-init"
adapter_id = "kosiasuzu/telos-llama3.1-8b-lora-adapter"

from telos.tokenizer import TelosTokenizer

tt = TelosTokenizer.from_pretrained(base_id)
base = AutoModelForCausalLM.from_pretrained(
    base_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)
model = PeftModel.from_pretrained(base, adapter_id)
model.eval()

# use tt + render/prompt as in the merged model card

CLI eval (PEFT mode):

telos eval-format-validity \
  --format telos \
  --model kosiasuzu/telos-agent-llama-3.1-8b-init \
  --adapter-mode peft \
  --adapter-id kosiasuzu/telos-llama3.1-8b-lora-adapter \
  --dataset kosiasuzu/telos-agent-trajectory-dataset \
  --split eval \
  --output results/telos_format_validity_peft.json

Training Details

Produced by telos train-telos-lora with --adapter-repo-id kosiasuzu/telos-llama3.1-8b-lora-adapter.

Full training narrative: see `telos-llama3.1-8b-lora-merged.md`.

Model Card Contact

• —Hugging Face: kosiasuzu