You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Gemma 105k Prompt Embedding Model

This model package was exported from the Stage 2 training run:

Run directory: 20260812_100911_gemma105k_matryoshka_v1
Export date: 2026-08-27

It is intended to be downloaded directly from Hugging Face and tested immediately by another team.

What Is Included

The full Sentence Transformers model at repo root
Prompt-aware configuration in config_sentence_transformers.json
Training/evaluation manifests under artifacts/
Post-train benchmark outputs under artifacts/post_eval/

Prompt Behavior

The exported model keeps the prompt prefixes used during training and evaluation:

Query prefix: task: search result | query:
Document prefix: title: none | text:

This means users can load the model normally:

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("your-org/gemma105k-prompt-embed")
emb = model.encode(
    ["what is photosynthesis?"],
    prompt_name="query",
)

Best Checkpoint Selection

load_best_model_at_end: True
metric_for_best_model: eval_DeepEdu_dot_ndcg@10
best_metric: 0.6407359510534348
best_model_checkpoint: output/20260812_100911_gemma105k_matryoshka_v1/checkpoints/checkpoint-200

Multi-Best Summary

{
  "deb": {
    "value": 0.6407359510534348,
    "step": 200,
    "checkpoint": "output/20260812_100911_gemma105k_matryoshka_v1/best_checkpoints/deb"
  },
  "rteb_financebench": {
    "value": 0.8054662482334365,
    "step": 200,
    "checkpoint": "output/20260812_100911_gemma105k_matryoshka_v1/best_checkpoints/rteb_financebench"
  },
  "rteb_humaneval": {
    "value": 0.987053136533919,
    "step": 800,
    "checkpoint": "output/20260812_100911_gemma105k_matryoshka_v1/best_checkpoints/rteb_humaneval"
  },
  "rteb_legalsummarization": {
    "value": 0.6590277899878882,
    "step": 400,
    "checkpoint": "output/20260812_100911_gemma105k_matryoshka_v1/best_checkpoints/rteb_legalsummarization"
  },
  "vnmteb_mini": {
    "value": 0.5384664016717299,
    "step": 200,
    "checkpoint": "output/20260812_100911_gemma105k_matryoshka_v1/best_checkpoints/vnmteb_mini"
  },
  "train_loss": {
    "value": 0.5513583866022251,
    "step": 800,
    "checkpoint": "output/20260812_100911_gemma105k_matryoshka_v1/best_checkpoints/train_loss"
  }
}

Post-Train Benchmark Summary

{
  "stage2_run": "20260812_100911_gemma105k_matryoshka_v1",
  "final_model_dir": "output/20260812_100911_gemma105k_matryoshka_v1/final_model",
  "benchmarks": {
    "sdg939q_canonical": {
      "status": "completed",
      "results_path": "output/20260812_100911_gemma105k_matryoshka_v1/post_eval/20260813_093214_sdg939q_canonical/eval_results.json",
      "comparison_path": "output/20260812_100911_gemma105k_matryoshka_v1/post_eval/20260813_093214_sdg939q_canonical/comparison.json",
      "wandb_run_url": "https://wandb.ai/aivforever/deepedu_embed/runs/hxzm1lvl"
    },
    "rteb_financebench": {
      "status": "completed",
      "results_path": "output/20260812_100911_gemma105k_matryoshka_v1/post_eval/20260813_101350_rteb_financebench/eval_results.json",
      "comparison_path": "output/20260812_100911_gemma105k_matryoshka_v1/post_eval/20260813_101350_rteb_financebench/comparison.json",
      "wandb_run_url": "https://wandb.ai/aivforever/deepedu_embed/runs/tyebh3p5"
    },
    "rteb_humaneval": {
      "status": "completed",
      "results_path": "output/20260812_100911_gemma105k_matryoshka_v1/post_eval/20260813_101421_rteb_humaneval/eval_results.json",
      "comparison_path": "output/20260812_100911_gemma105k_matryoshka_v1/post_eval/20260813_101421_rteb_humaneval/comparison.json",
      "wandb_run_url": "https://wandb.ai/aivforever/deepedu_embed/runs/frumsd6y"
    },
    "rteb_legalsummarization": {
      "status": "completed",
      "results_path": "output/20260812_100911_gemma105k_matryoshka_v1/post_eval/20260813_101507_rteb_legalsummarization/eval_results.json",
      "comparison_path": "output/20260812_100911_gemma105k_matryoshka_v1/post_eval/20260813_101507_rteb_legalsummarization/comparison.json",
      "wandb_run_url": "https://wandb.ai/aivforever/deepedu_embed/runs/zev4hay1"
    },
    "vnmteb10": {
      "status": "completed",
      "results_path": "output/20260812_100911_gemma105k_matryoshka_v1/post_eval/20260813_101635_vnmteb10/eval_results.json",
      "comparison_path": "output/20260812_100911_gemma105k_matryoshka_v1/post_eval/20260813_101635_vnmteb10/comparison.json",
      "wandb_run_url": "https://wandb.ai/aivforever/deepedu_embed/runs/e8defs9f"
    }
  },
  "created_at": "2026-08-13T18:08:22.182796"
}

Artifact Layout

root: loadable Sentence Transformers model
artifacts/stage2_manifest.json: training/export manifest
artifacts/best_summary.json: best-checkpoint summary
artifacts/post_eval_summary.json: post-train benchmark summary
artifacts/post_eval/: raw post-eval outputs

Minimal Load Example

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("your-org/gemma105k-prompt-embed")
query_vec = model.encode(
    ["task: search result | query: what is photosynthesis?"],
    normalize_embeddings=True,
)

Downloads last month: 9

Safetensors

Model size

0.3B params

Tensor type

BF16