oddadmix/50M-Egyptian-Translation-v1

50M-Egyptian-Translation-v1 — English → Egyptian Arabic

A 51.8M-parameter small language model finetuned to translate English into Egyptian colloquial Arabic (اللهجة المصرية العامية). It is a supervised finetune of `oddadmix/50M-2048-Emhotob`, a tiny Arabic base model trained from scratch.

Despite its size, the model produces natural, dialect-correct Egyptian Arabic on conversational text.

Evaluation

Evaluated on a deterministic held-out set of 3,000 pairs (seed=42), decoded greedily (do_sample=False), scored with sacreBLEU:

The saved weights are the best checkpoint by validation loss (eval_loss=1.271, epoch 2 of 3).

Example translations

Real greedy-decoded outputs from the held-out set (English → model output):

A larger set of 20 examples (with references) is included in `eval_translation.json`.

Usage

The model uses a ChatML prompt format with a fixed system instruction.

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "oddadmix/50M-Egyptian-Translation-v1"
tok = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, dtype=torch.bfloat16).to("cuda").eval()

SYSTEM = "أنت مترجم محترف. ترجم النص الإنجليزي إلى اللهجة المصرية العامية."

def translate(english: str) -> str:
    prompt = (
        f"<|im_start|>system\n{SYSTEM}<|im_end|>\n"
        f"<|im_start|>user\n{english.strip()}<|im_end|>\n"
        f"<|im_start|>assistant\n"
    )
    ids = tok(prompt, return_tensors="pt", add_special_tokens=False).to(model.device)
    if tok.bos_token_id is not None:  # training prepends BOS
        bos = torch.tensor([[tok.bos_token_id]], device=model.device)
        ids["input_ids"] = torch.cat([bos, ids["input_ids"]], dim=1)
        ids["attention_mask"] = torch.cat([torch.ones_like(bos), ids["attention_mask"]], dim=1)
    out = model.generate(**ids, max_new_tokens=256, do_sample=False,
                         eos_token_id=tok.eos_token_id, pad_token_id=tok.pad_token_id)
    return tok.decode(out[0, ids["input_ids"].size(1):], skip_special_tokens=True).strip()

print(translate("Man, these things happen. Sometimes Liverpool loses the match."))
# → يا عم الحاجات دي بتحصل. ساعات ليفربول يخسر الماتش.

Training

• —Base model: oddadmix/50M-2048-Emhotob (Llama arch, ~51.8M params)
• —Dataset: oddadmix/egyptian-translation-dataset-2.9-openai-batch (135K English/Egyptian-Arabic pairs)
• —Method: HuggingFace Trainer, ChatML format, prompt-masked cross-entropy (loss only on the Arabic assistant turn). Two ChatML special tokens (<|im_start|>, <|im_end|>) were added and the embeddings resized.
• —Hyperparameters: 3 epochs · effective batch 64 · LR 3e-4 (cosine, 5% warmup) · bf16 · max length 1024 · load_best_model_at_end on eval_loss.
• —Split: 132,282 train / 3,000 deterministic held-out eval (seed=42).

<!-- BEGIN: leaderboard-en2egy-eval -->

Out-of-domain evaluation — Egyptian Arabic Translation Benchmark

The results above are in-domain: a held-out split of the same corpus this model was trained on. The numbers below are out-of-domain — the same model scored on the Egyptian Arabic Translation Benchmark (oddadmix/egyptian-arabic-translation-benchmark, 319 English→Egyptian pairs written by a different annotator with different orthographic conventions).

Expect these to be substantially lower than the in-domain scores. That gap is the generalization penalty, not a regression — both numbers are real, they measure different things.

Decoding is deterministic greedy (do_sample=False, no repetition penalty), using the exact ChatML prompt format the model was trained with — the same protocol as every other number in this study.

Where this rung sits

Reading these numbers

BLEU understates quality on this set. Scoring is against a single reference, so a correct translation that picks a different valid word is penalized — e.g. فريش vs the reference's طازة for "fresh", or التليفون اللي ضاع vs تليفونها الضايع for "her lost phone". Both are good Egyptian; only one matches the reference. chrF and METEOR track perceived quality more closely here.

At 319 rows, differences of roughly 1–2 BLEU between adjacent rungs are within noise.

These are small models — 5M to 50M parameters, orders of magnitude below the large systems typically evaluated on this benchmark. The result of interest is the scaling curve and per-parameter efficiency, not absolute rank against models 100–1000× the size. <!-- END: leaderboard-en2egy-eval -->

Limitations

• —A 50M model: expect errors on rare / technical vocabulary and occasional repetition loops on longer generations (mitigate with repetition_penalty and no_repeat_ngram_size at inference).
• —Trained on conversational Egyptian; other Arabic dialects or formal MSA are out of scope.
• —Gender is disambiguated only from context, so it may default to masculine for ambiguous English inputs.

License

Apache-2.0, inherited from the base model.