A 25.3M-parameter model that translates both ways between English and Modern Standard Arabic (الفصحى). A single set of weights serves both directions; a direction-specific system prompt selects which way to translate.

Finetuned from oddadmix/Emhotob-25M, a tiny Llama-architecture base (hidden 384, 8 layers, 6 heads, vocab 32000, tied embeddings).

Scaling study. This is one rung of a from-scratch Arabic scaling study that runs an identical SFT + eval recipe across bases from 0.5M to 50M parameters to locate where translation emerges. On the headline MSA↔Egyptian pair, output is degenerate at ≤1M, becomes real-but-rough at 5M, and usable at 10M+. See the sibling oddadmix/50M-English-MSA-v1 for the fluent reference.

Evaluation

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

Direction	sacreBLEU	chrF
English → MSA	40.65	62.42
MSA → English	44.70	60.68

Saved weights are the best checkpoint by validation loss (eval_loss = 0.703). 20 samples per direction with references are in eval_bidirectional.json.

Example translations

Real greedy-decoded outputs from the held-out set:

English → MSA

Source	Model output	Reference
I'm just trying to reassure myself. What's the worst that could happen? It's just mentally exhausting, but they're stand	أنا فقط أحاول طمأنة نفسي. ما الأسوأ أن يحدث ذلك؟ إنه مرهق عقليًا، لكنهم يقفون بقوة. حقًا، إنهم يقفون بقوة.	أنا فقط أحاول أن أطمئن نفسي. ما هو الأسوأ الذي يمكن أن يحدث؟ إنه مرهق عقليًا فقط، لكنهم يقفون بحزم. حقًا، إنهم يقفون بحز
Thank you very much, you are so kind.	شكرًا جزيلًا لك، أنت لطيف جدًا.	شكرًا جزيلًا لك، أنت لطيف للغاية.

MSA → English

Source	Model output	Reference
أنا فقط أحاول أن أطمئن نفسي. ما هو الأسوأ الذي يمكن أن يحدث؟ إنه مرهق عقليًا فقط، لكنهم يقفون بحزم. حقًا، إنهم يقفون بحز	I'm just trying to reassure myself. What's the worst can happen? It's just mentally exhausted, but they're standing down	I'm just trying to reassure myself. What's the worst that could happen? It's just mentally exhausting, but they're stand
شكرًا جزيلًا لك، أنت لطيف للغاية.	Thank you so much, you're so sweet.	Thank you very much, you are so kind.

Usage

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

python

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "oddadmix/Emhotob-25M-English-MSA-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 اللغة العربية الفصحى."

def translate(text, system=SYSTEM):
    prompt = (f"<|im_start|>system\n{system}<|im_end|>\n"
              f"<|im_start|>user\n{text.strip()}<|im_end|>\n<|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:
        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()

Training

Base model: oddadmix/Emhotob-25M (Llama arch, hidden 384, 8 layers, 6 heads, vocab 32000, tied embeddings; 25,271,424 params after resizing for 2 ChatML tokens)
Dataset: oddadmix/egyptian-msa-2.9-openai-bytedance-translations
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).
Hyperparameters: 3 epochs · effective batch 64 · LR 3e-4 (cosine, 5% warmup) · bf16 · max length 1024 · load_best_model_at_end on eval_loss.
Eval split: 3,000 deterministic held-out pairs (seed=42), scored both directions.

Limitations

A ~25.3M model: reliable on short/common sentences, but drift, repetition, and errors appear on long or rare inputs. Gender is disambiguated only from context. For fluent translation use the 50M sibling.

License

Apache-2.0, inherited from the base model.