tarekziade HF Staff
Publish MIC usage, slim fixture and reviewed Transformers expectations
d7fe47d verified |
Download integrations/transformers/usage.md from huggingface-mic/Hunyuan-0.5B-Instruct-mic: direct link, hf CLI and curl.
- Browser
- Download file 3.67 kB
-
https://huggingface.co/huggingface-mic/Hunyuan-0.5B-Instruct-mic/resolve/main/integrations/transformers/usage.md
- Command line
-
hf download hf://huggingface-mic/Hunyuan-0.5B-Instruct-mic/integrations/transformers/usage.md
-
curl -L -o usage.md https://huggingface.co/huggingface-mic/Hunyuan-0.5B-Instruct-mic/resolve/main/integrations/transformers/usage.md
3.67 kB
Hunyuan-0.5B-Instruct: causal LM with Transformers
This page is the Transformers contract of the Hunyuan-0.5B-Instruct MIC and also a usage example. doc-builder renders the runnable blocks below as documentation, and its pytest plugin runs them as written against tencent/Hunyuan-0.5B-Instruct. The MIC runner runs the same blocks against the slim fixture and the checkpoint in integration.json. Lines marked # doc-builder: hide only matter there: they swap in the pinned checkpoint or variant and print the result record.
Predict the next token
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, set_seed
import json # doc-builder: hide
import os # doc-builder: hide
model_id = "tencent/Hunyuan-0.5B-Instruct"
model_id = os.environ.get("MIC_MODEL", model_id) # doc-builder: hide
variant = json.loads(os.environ.get("MIC_MODEL_KWARGS", "{}")) # doc-builder: hide
device = "cuda" if torch.cuda.is_available() else "cpu"
set_seed(0) # seeds torch, NumPy, and Python: loading, dropout, and data augmentation draw random numbers
tokenizer = AutoTokenizer.from_pretrained(model_id)
if tokenizer.pad_token is None: # Hunyuan defines no pad token; left padding keeps the last position real
tokenizer.pad_token = tokenizer.eos_token
tokenizer.padding_side = "left"
model = AutoModelForCausalLM.from_pretrained(model_id).to(device)
if variant: # doc-builder: hide
del model # doc-builder: hide
model = AutoModelForCausalLM.from_pretrained(model_id, **variant).to(device) # doc-builder: hide
prompts = [
"The capital of France is",
"Write a short summary of the benefits of regular exercise:",
]
inputs = tokenizer(prompts, return_tensors="pt", padding=True).to(device)
with torch.no_grad():
logits = model(**inputs).logits
# left padding: the last position is the real next-token position
last = logits[:, -1, :]
next_ids = last.argmax(dim=-1)
print(tokenizer.batch_decode(last.argmax(dim=-1, keepdim=True)))
One training step
# causal-LM labels: copy input_ids and mask padded positions
labels = inputs["input_ids"].clone()
labels[inputs["attention_mask"] == 0] = -100
model.train()
optimizer = torch.optim.SGD(model.parameters(), lr=0.01)
loss = model(**inputs, labels=labels).loss
loss.backward()
before = {n: p.detach().cpu().clone() for n, p in model.named_parameters() if p.requires_grad} # doc-builder: hide
optimizer.step()
print(f"loss: {loss.item():.4f}")
if os.environ.get("MIC_RUNNER"): # doc-builder: hide
named = dict(model.named_parameters())
checked = (
"model.layers.0.self_attn.q_proj.weight",
"model.layers.0.mlp.gate_proj.weight",
)
result = {
"runtime": {
"device": model.device.type,
"dtype": str(model.dtype).removeprefix("torch."),
"attn_implementation": model.config._attn_implementation,
},
"next_tokens": next_ids.tolist(),
"next_text": tokenizer.batch_decode(next_ids, skip_special_tokens=True),
"probe_logits": last[:, :4].float().tolist(),
"training": {
"loss": loss.item(),
"gradients": {
n: bool(named[n].grad is not None and torch.count_nonzero(named[n].grad) > 0 and torch.isfinite(named[n].grad).all())
for n in checked
},
"updated": any(not torch.equal(p.detach().cpu(), before[n]) for n, p in model.named_parameters() if n in before),
},
}
print(json.dumps(result))