p5-ann

p5-ann

Full Transformers safetensors model based on microsoft/Phi-4-mini-instruct.

Notes

• This repo is a merged full model export, not an adapter-only repo.
• annihilate_merge_metadata.json records the source adapter and merge details.
• The repo intentionally uses the built-in Transformers Phi3/Phi loader path. Stale custom remote-code files were removed so trust_remote_code=True and False both resolve to the built-in loader.
• Generation should stop on <|end|> / <|endoftext|>; generation_config.json contains both stop IDs.

Quick Load

from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "tjcrims0n/p5-ann"
tokenizer = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(repo, device_map="auto")

messages = [
    {"role": "system", "content": "Answer directly and concisely."},
    {"role": "user", "content": "Hello."},
]
inputs = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True).to(model.device)
outputs = model.generate(
    inputs,
    max_new_tokens=80,
    eos_token_id=model.generation_config.eos_token_id,
    pad_token_id=model.generation_config.pad_token_id,
)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Smoke Test Notes

Local verification passed for greeting, arithmetic, and code prompts after removing stale remote-code hooks from config.json.