Emhotob-5M-MSA-Egyptian-v1

Emhotob-5M-MSA-Egyptian-v1 — Bidirectional MSA ↔ Egyptian Arabic (~5M params)

A 5.08M-parameter model that translates both ways between Modern Standard Arabic (الفصحى) and Egyptian colloquial Arabic (المصرية العامية). A single set of weights serves both directions; a direction-specific Arabic system prompt selects which way to translate.

Finetuned from oddadmix/Emhotob-5M, a tiny Llama-architecture base (hidden size 128, 5 layers, 4 heads, tied embeddings).

Where translation "turns on." This model is part of a scaling study that runs the exact SFT + evaluation recipe of oddadmix/50M-MSA-Egyptian-v1 across shrinking base models. At 500K and 1M params the model collapses to a degenerate repeated token (BLEU ≈ 0). At 5M it genuinely translates — roughly, with drift and repetition, but with correct register-switching and real lexical choices. It is a working demonstration, not a production translator; for fluent output use the 50M sibling.

Evaluation

Evaluated on a deterministic held-out set of 3,000 pairs (seed=42), decoded greedily (do_sample=False, no repetition penalty), scored with sacreBLEU:

The saved weights are the best checkpoint by validation loss (eval_loss = 3.552, epoch 3 of 3; loss keeps improving across all three epochs, unlike the sub-2M bases which plateau immediately).

Scaling context (same data, recipe, and eval; only the base model changes)

Example translations

Real greedy-decoded outputs from the held-out set (rough — this is a 5M model):

MSA → Egyptian

Egyptian → MSA

The model reliably switches register (Egyptian markers بس/إيه/هقولك one way, MSA فقط/ما هي the other) but frequently drifts or repeats on longer inputs. 20 samples per direction with references are in eval_bidirectional.json.

Usage

ChatML format. Pick the system prompt for the direction you want:

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "oddadmix/Emhotob-5M-MSA-Egyptian-v1"
tok = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, dtype=torch.bfloat16).to("cuda").eval()

SYS_TO_EGY = "أنت مترجم محترف. ترجم النص من اللغة العربية الفصحى إلى اللهجة المصرية العامية."
SYS_TO_MSA = "أنت مترجم محترف. ترجم النص من اللهجة المصرية العامية إلى اللغة العربية الفصحى."

def translate(text: str, system: str) -> str:
    prompt = (
        f"<|im_start|>system\n{system}<|im_end|>\n"
        f"<|im_start|>user\n{text.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("شكرًا جزيلًا لك، أنت لطيف للغاية.", SYS_TO_EGY))

Training

• Base model: oddadmix/Emhotob-5M (Llama arch, hidden 128, 5 layers, 4 heads, vocab 32000, tied embeddings; 5,080,704 params after resizing for 2 ChatML tokens)
• Dataset: oddadmix/egyptian-msa-2.9-openai-bytedance-translations (132K rows, egyptian/msa columns)
• Method: HuggingFace Trainer, ChatML, prompt-masked cross-entropy (loss only on the assistant turn). Each row is exploded into two training examples (one per direction). Two ChatML special tokens (<|im_start|>, <|im_end|>) were added and 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: 129,009 train / 3,000 deterministic held-out (seed=42), scored both directions.

Limitations

• A 5M model at the emergence threshold for this task: expect frequent drift, repetition, hallucinated content, and errors on anything beyond short everyday sentences.
• Gender is disambiguated only from context; ambiguous inputs may default one way.
• For fluent translation use oddadmix/50M-MSA-Egyptian-v1. For the degenerate smaller points in this study see oddadmix/Emhotob-500K-MSA-Egyptian-v1.

License

Apache-2.0, inherited from the base model.