Text Classification
Transformers
Safetensors
English
modernbert
commitment-detection
decision-detection
experimental
vela
text-embeddings-inference
Instructions to use TwilightTechie/vela-commitment-classifier-307m with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use TwilightTechie/vela-commitment-classifier-307m with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-classification", model="TwilightTechie/vela-commitment-classifier-307m")# pip install -U transformers accelerate # Load model directly from transformers import AutoTokenizer, AutoModelForSequenceClassification tokenizer = AutoTokenizer.from_pretrained("TwilightTechie/vela-commitment-classifier-307m") model = AutoModelForSequenceClassification.from_pretrained("TwilightTechie/vela-commitment-classifier-307m", device_map="auto") - Notebooks
- Google Colab
- Kaggle
File size: 3,428 Bytes
f0e7192 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 | """HF standard sequence checkpoints with an unchanged label contract."""
# Heavy imports stay optional for data-contract tests.
# ruff: noqa: PLC0415
import json
from pathlib import Path
from .data import load_contract
from .task_head import verify_saved_task_head
def configuration_overrides(contract):
"""Validate optional standard HF task settings without guessing a pooling mode."""
content = json.loads(Path(contract).read_text())
problem_type = content.get("problem_type", "single_label_classification")
if problem_type not in {
"single_label_classification",
"multi_label_classification",
}:
raise ValueError("Unsupported classification problem type")
overrides = {"problem_type": problem_type}
if "classifier_pooling" in content:
pooling = content["classifier_pooling"]
if pooling not in {"mean", "cls"}:
raise ValueError("Sequence pooling must be explicitly mean or cls")
overrides["classifier_pooling"] = pooling
return overrides
def load_task_config(base, **overrides):
"""Keep optional execution settings on the configuration that owns them."""
from transformers import AutoConfig
config = AutoConfig.from_pretrained(base, **overrides)
if hasattr(config, "reference_compile"):
config.reference_compile = False
return config
def load_model(base, contract, adapter=None, trainable=False):
import torch
from transformers import AutoModelForSequenceClassification, AutoTokenizer
label_to_id, id_to_label = load_contract(contract)
overrides = configuration_overrides(contract)
tokenizer = AutoTokenizer.from_pretrained(base)
config = load_task_config(
base,
num_labels=len(label_to_id),
label2id=label_to_id,
id2label=id_to_label,
**overrides,
)
model, loading = AutoModelForSequenceClassification.from_pretrained(
base,
config=config,
torch_dtype=torch.float32,
attn_implementation="sdpa",
output_loading_info=True,
)
missing_encoder = [
name for name in loading["missing_keys"] if name.startswith("model.")
]
if missing_encoder:
raise ValueError(f"Checkpoint is missing encoder tensors: {missing_encoder}")
if adapter:
from peft import PeftModel
model = PeftModel.from_pretrained(model, adapter, is_trainable=trainable)
elif trainable:
raise ValueError("Continuation requires an explicit initial adapter")
if model.config.label2id != label_to_id:
raise ValueError("The loaded classifier changed the label contract")
return model, tokenizer, label_to_id, id_to_label
def save_adapter(model, tokenizer, destination, base_id, base_revision):
import json
destination.mkdir(parents=True, exist_ok=True)
model.peft_config["default"].base_model_name_or_path = base_id
model.peft_config["default"].revision = base_revision
model.save_pretrained(destination)
scope = verify_saved_task_head(model, destination / "adapter_model.safetensors")
(destination / "task-head.json").write_text(json.dumps(scope, indent=2) + "\n")
tokenizer.save_pretrained(destination)
(destination / "label_mapping.json").write_text(
json.dumps(
{"label2id": model.config.label2id, "id2label": model.config.id2label},
indent=2,
)
+ "\n"
)
|