You can deploy this model and start earning money today!

aamingem/
texformer-100m-int8_weight_only

Name: texformer-100m-int8_weight_only
Availability: InStock

Views

No views yet

texformer-100m-int8_weight_only

This repository contains a TeXformer 100m checkpoint in int8_weight_only precision for OCR-to-LaTeX generation.

This is a custom TeXformer architecture checkpoint (model.pt) plus tokenizer assets. It is not a standard transformers AutoModel checkpoint. This export is derived from a checkpoint trained in bf16.

Files

model.pt: TeXformer checkpoint
tokenizer/pdf_tokenizer.json: PDF-side tokenizer
tokenizer/latex_tokenizer.json: LaTeX-side tokenizer
tokenizer/pdf_tags.json: frequent PDF tag metadata
tokenizer/latex_commands.json: frequent LaTeX command metadata

Architecture

Parameters (deduplicated): 96,354,304
Parameters (state_dict entries): 120,930,304
Encoder layers: 6
Decoder layers: 6
Hidden size (d_model): 512
Attention heads: 8
Feed-forward size (d_ff): 2304
Max encoder length: 2560
Max decoder length: 2560
Stored precision: int8_weight_only

Quantization

Quantization method: int8_weight_only_pc
Checkpoint payload key: quantized_state_dict
Runtime support: dequantize to float first, then run on CPU/CUDA/MPS.
CUDA/MPS: supported only after dequantization (not true INT8 execution).
Original model training precision: bf16
Notes: Dequantize to float tensors before loading into standard TeXFormer.

Usage

python

from pathlib import Path
import torch
from huggingface_hub import snapshot_download
from texformer.models.model import TeXFormer, TeXFormerConfig
from texformer.models.quantize import dequantize_weight_only_state_dict
from texformer.tokenization.tokenizer import TeXFormerTokenizer

repo_id = "aamingem/texformer-100m-int8_weight_only"
local_dir = Path(snapshot_download(repo_id=repo_id))
tokenizer_dir = local_dir / "tokenizer"
checkpoint = torch.load(local_dir / "model.pt", map_location="cpu", weights_only=False)

if torch.cuda.is_available():
    device = torch.device("cuda")
elif hasattr(torch.backends, "mps") and torch.backends.mps.is_available():
    device = torch.device("mps")
else:
    device = torch.device("cpu")

config = TeXFormerConfig(**checkpoint["config"])
state_dict = dequantize_weight_only_state_dict(checkpoint["quantized_state_dict"])
model = TeXFormer(config)
model.load_state_dict(state_dict, strict=False)
model = model.to(device)

tokenizer = TeXFormerTokenizer(tokenizer_dir)
print("Loaded dequantized model with tokenizer:", tokenizer.pdf_vocab_size, tokenizer.latex_vocab_size)

Intended Use

OCR-to-LaTeX / PDF-text-to-LaTeX sequence generation
Research and experimentation on scientific document conversion

Limitations

May produce incorrect or non-compiling LaTeX.
Performance depends on input extraction quality.
Not intended for high-stakes use without human verification.