e5-base-hebrew-qa-v2-myd-r1

MYD Round-1 Hebrew fine-tune of intfloat/multilingual-e5-base on 15,142 persona↔CEO Q-A pairs extracted from the MYD synthetic dialog panel (CEOs: Rafael, Adel, Antonio).

Usage

from sentence_transformers import SentenceTransformer
model = SentenceTransformer("oridror/e5-base-hebrew-qa-v2-myd-r1")
queries  = ["query: מה זה MedBed?"]
passages = ["passage: MedBed הוא פרויקט לריפוי הוליסטי..."]
q_emb = model.encode(queries, normalize_embeddings=True)
p_emb = model.encode(passages, normalize_embeddings=True)
sim = (q_emb @ p_emb.T)[0][0]
print(sim)

Important: This model uses the E5-family prefix convention — always prepend query: to queries and passage: to documents at inference time. Forgetting the prefix will silently degrade quality.

Eval (n=500 held-out Hebrew Q-A pairs)

Metric	Value
Accuracy@1	0.722
Accuracy@5	0.852
MRR@10	0.7774

Training

• Base: intfloat/multilingual-e5-base
• Loss: MultipleNegativesRankingLoss (in-batch negatives, scale=20)
• Epochs: 2
• LR: 2e-5 with 100 warmup steps
• Hardware: NVIDIA A100 80GB PCIe (RunPod)
• Run: myd-r1-runpod-5-models (2026-04-22)

Data

Extracted from 6-ai/synthetic-panel/output/dialogs/all_dialogs.jsonl — 3,925 generated Hebrew dialogs (persona + CEO turns). Every persona-role turn paired with the immediately-following CEO-role turn yielded 15,642 Q-A pairs. Split: 15,142 train / 500 eval (seed 42).

License

Apache-2.0 (inherited from base model).

Part of MYD

This model is part of the MYD generative-system stack. Routed via myd-router policy as embed.he candidate.