Instructions to use od2961/maxent-grpo-models with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use od2961/maxent-grpo-models with Transformers:
# pip install -U transformers accelerate # Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("od2961/maxent-grpo-models", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Index 476 verified experimental models
Browse files- README.md +3 -3
- catalog.json +74 -2
- experiments/E112-R1/README.md +17 -8
README.md
CHANGED
|
@@ -8,7 +8,7 @@ tags:
|
|
| 8 |
|
| 9 |
Terminal fine-tuned models for studying correctness and verified output support. Browse by study, base model, domain, method and seed; restore each export from its recorded immutable revision.
|
| 10 |
|
| 11 |
-
**Paper artifact collection ·
|
| 12 |
|
| 13 |
The fixed archival selection contains admitted completed models from the source studies listed below. Counts describe archive availability, not live training progress or a replacement for the papers’ frozen analysis sets.
|
| 14 |
|
|
@@ -39,14 +39,14 @@ The broader paper collection is being expanded to additional baselines, ablation
|
|
| 39 |
| [E86](experiments/E86/README.md) | Fixed semantic without replay · Falcon3-1B | 0 / 25 |
|
| 40 |
| [E87](experiments/E87/README.md) | Historical fixed-semantic probe · Qwen2.5-3B | 5 / 5 |
|
| 41 |
| [E109](experiments/E109/README.md) | Repaired Python replay comparators | 5 / 10 |
|
| 42 |
-
| [E112-R1](experiments/E112-R1/README.md) | Verified-support semantic + proposal bundle |
|
| 43 |
| [E121](experiments/E121/README.md) | Fixed-bank exemplar survival | 0 / 5 |
|
| 44 |
|
| 45 |
## Artifact availability
|
| 46 |
|
| 47 |
| Scientific records selected | Deployable model exports | Verified model exports | Model weights unavailable |
|
| 48 |
| ---: | ---: | ---: | ---: |
|
| 49 |
-
| 762 | 707 |
|
| 50 |
|
| 51 |
Model weights are unavailable for 55 retained scientific records: E95 (55). Their metadata and admitted results remain part of the research record; they are excluded from the downloadable-model catalog.
|
| 52 |
|
|
|
|
| 8 |
|
| 9 |
Terminal fine-tuned models for studying correctness and verified output support. Browse by study, base model, domain, method and seed; restore each export from its recorded immutable revision.
|
| 10 |
|
| 11 |
+
**Paper artifact collection · 476 / 707 selected exports verified. Transfer is in progress.**
|
| 12 |
|
| 13 |
The fixed archival selection contains admitted completed models from the source studies listed below. Counts describe archive availability, not live training progress or a replacement for the papers’ frozen analysis sets.
|
| 14 |
|
|
|
|
| 39 |
| [E86](experiments/E86/README.md) | Fixed semantic without replay · Falcon3-1B | 0 / 25 |
|
| 40 |
| [E87](experiments/E87/README.md) | Historical fixed-semantic probe · Qwen2.5-3B | 5 / 5 |
|
| 41 |
| [E109](experiments/E109/README.md) | Repaired Python replay comparators | 5 / 10 |
|
| 42 |
+
| [E112-R1](experiments/E112-R1/README.md) | Verified-support semantic + proposal bundle | 6 / 49 |
|
| 43 |
| [E121](experiments/E121/README.md) | Fixed-bank exemplar survival | 0 / 5 |
|
| 44 |
|
| 45 |
## Artifact availability
|
| 46 |
|
| 47 |
| Scientific records selected | Deployable model exports | Verified model exports | Model weights unavailable |
|
| 48 |
| ---: | ---: | ---: | ---: |
|
| 49 |
+
| 762 | 707 | 476 | 55 |
|
| 50 |
|
| 51 |
Model weights are unavailable for 55 retained scientific records: E95 (55). Their metadata and admitted results remain part of the research record; they are excluded from the downloadable-model catalog.
|
| 52 |
|
catalog.json
CHANGED
|
@@ -1,9 +1,9 @@
|
|
| 1 |
{
|
| 2 |
"schema": "experiment-model-archive-catalog-v1",
|
| 3 |
-
"updated_at_utc": "2026-09-11T18:
|
| 4 |
"repo_id": "od2961/maxent-grpo-models",
|
| 5 |
"expected_model_count": 707,
|
| 6 |
-
"verified_model_count":
|
| 7 |
"models": [
|
| 8 |
{
|
| 9 |
"experiment": "E100",
|
|
@@ -329,6 +329,78 @@
|
|
| 329 |
"local_weights_retired": true,
|
| 330 |
"export_bytes": 3348983622
|
| 331 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 332 |
{
|
| 333 |
"experiment": "E114",
|
| 334 |
"model": "Qwen2.5-3B-Instruct",
|
|
|
|
| 1 |
{
|
| 2 |
"schema": "experiment-model-archive-catalog-v1",
|
| 3 |
+
"updated_at_utc": "2026-09-11T18:33:00.256408+00:00",
|
| 4 |
"repo_id": "od2961/maxent-grpo-models",
|
| 5 |
"expected_model_count": 707,
|
| 6 |
+
"verified_model_count": 476,
|
| 7 |
"models": [
|
| 8 |
{
|
| 9 |
"experiment": "E100",
|
|
|
|
| 329 |
"local_weights_retired": true,
|
| 330 |
"export_bytes": 3348983622
|
| 331 |
},
|
| 332 |
+
{
|
| 333 |
+
"experiment": "E112-R1",
|
| 334 |
+
"model": "Falcon3-1B-Instruct",
|
| 335 |
+
"domain": "countdown",
|
| 336 |
+
"method": "verified_support_discovery",
|
| 337 |
+
"seed": 55,
|
| 338 |
+
"step": 3073,
|
| 339 |
+
"repo_prefix": "experiments/E112-R1/Falcon3-1B-Instruct/countdown/verified_support_discovery/seed-55/step-03073",
|
| 340 |
+
"commit_sha": "84e784c10b904062a75bdfd9c854b96a6a79df4d",
|
| 341 |
+
"local_weights_retired": true,
|
| 342 |
+
"export_bytes": 3348983622
|
| 343 |
+
},
|
| 344 |
+
{
|
| 345 |
+
"experiment": "E112-R1",
|
| 346 |
+
"model": "Falcon3-1B-Instruct",
|
| 347 |
+
"domain": "countdown",
|
| 348 |
+
"method": "verified_support_discovery",
|
| 349 |
+
"seed": 56,
|
| 350 |
+
"step": 3073,
|
| 351 |
+
"repo_prefix": "experiments/E112-R1/Falcon3-1B-Instruct/countdown/verified_support_discovery/seed-56/step-03073",
|
| 352 |
+
"commit_sha": "240d248dbbb6cfa8b5dbd989de77e9f9b71b63a3",
|
| 353 |
+
"local_weights_retired": true,
|
| 354 |
+
"export_bytes": 3348983622
|
| 355 |
+
},
|
| 356 |
+
{
|
| 357 |
+
"experiment": "E112-R1",
|
| 358 |
+
"model": "Falcon3-1B-Instruct",
|
| 359 |
+
"domain": "countdown",
|
| 360 |
+
"method": "verified_support_discovery",
|
| 361 |
+
"seed": 57,
|
| 362 |
+
"step": 3073,
|
| 363 |
+
"repo_prefix": "experiments/E112-R1/Falcon3-1B-Instruct/countdown/verified_support_discovery/seed-57/step-03073",
|
| 364 |
+
"commit_sha": "60c7d9f383912e1969f5ecf5a38e15e998fe71a2",
|
| 365 |
+
"local_weights_retired": true,
|
| 366 |
+
"export_bytes": 3348983622
|
| 367 |
+
},
|
| 368 |
+
{
|
| 369 |
+
"experiment": "E112-R1",
|
| 370 |
+
"model": "Falcon3-1B-Instruct",
|
| 371 |
+
"domain": "countdown",
|
| 372 |
+
"method": "verified_support_discovery",
|
| 373 |
+
"seed": 58,
|
| 374 |
+
"step": 3073,
|
| 375 |
+
"repo_prefix": "experiments/E112-R1/Falcon3-1B-Instruct/countdown/verified_support_discovery/seed-58/step-03073",
|
| 376 |
+
"commit_sha": "27c3efc0ab3bf56284c60fd3965edf5c0a3e23a9",
|
| 377 |
+
"local_weights_retired": true,
|
| 378 |
+
"export_bytes": 3348983622
|
| 379 |
+
},
|
| 380 |
+
{
|
| 381 |
+
"experiment": "E112-R1",
|
| 382 |
+
"model": "Falcon3-1B-Instruct",
|
| 383 |
+
"domain": "graph_coloring",
|
| 384 |
+
"method": "verified_support_discovery",
|
| 385 |
+
"seed": 55,
|
| 386 |
+
"step": 3073,
|
| 387 |
+
"repo_prefix": "experiments/E112-R1/Falcon3-1B-Instruct/graph_coloring/verified_support_discovery/seed-55/step-03073",
|
| 388 |
+
"commit_sha": "049be19138fb25ceb3f2178ad89c48937f333353",
|
| 389 |
+
"local_weights_retired": true,
|
| 390 |
+
"export_bytes": 3348983622
|
| 391 |
+
},
|
| 392 |
+
{
|
| 393 |
+
"experiment": "E112-R1",
|
| 394 |
+
"model": "Falcon3-1B-Instruct",
|
| 395 |
+
"domain": "graph_coloring",
|
| 396 |
+
"method": "verified_support_discovery",
|
| 397 |
+
"seed": 56,
|
| 398 |
+
"step": 3073,
|
| 399 |
+
"repo_prefix": "experiments/E112-R1/Falcon3-1B-Instruct/graph_coloring/verified_support_discovery/seed-56/step-03073",
|
| 400 |
+
"commit_sha": "15e2f9cc9b4822cf6120dcd61500e7bb7a634a89",
|
| 401 |
+
"local_weights_retired": true,
|
| 402 |
+
"export_bytes": 3348983622
|
| 403 |
+
},
|
| 404 |
{
|
| 405 |
"experiment": "E114",
|
| 406 |
"model": "Qwen2.5-3B-Instruct",
|
experiments/E112-R1/README.md
CHANGED
|
@@ -18,14 +18,14 @@ These are terminal exports from the registered eight-pass training horizon. The
|
|
| 18 |
|
| 19 |
| Registered new training cells | Scientific records selected | Deployable exports selected | Verified exports |
|
| 20 |
| ---: | ---: | ---: | ---: |
|
| 21 |
-
| 75 | 49 | 49 |
|
| 22 |
|
| 23 |
Scientific records preserve the admitted result identity. Deployable exports additionally have available model weights; verified exports have completed remote verification. Cells outside the fixed selection are not silently counted, and unverified model folders are not linked below.
|
| 24 |
|
| 25 |
| Model | Verified / selected | Seeds in selection |
|
| 26 |
| --- | ---: | --- |
|
| 27 |
| [Qwen2.5-0.5B-Instruct](#model-qwen05b) | 0 / 25 | 43, 44, 45, 46, 47 |
|
| 28 |
-
| [Falcon3-1B-Instruct](#model-falcon1b) |
|
| 29 |
|
| 30 |
## Methods and collections
|
| 31 |
|
|
@@ -34,7 +34,8 @@ Scientific records preserve the admitted result identity. Deployable exports add
|
|
| 34 |
|
| 35 |
ReplayDr.GRPO augmented with semantic regularization and an isolated verified proposal sampler; a bundled comparison.
|
| 36 |
|
| 37 |
-
|
|
|
|
| 38 |
|
| 39 |
<a id="models"></a>
|
| 40 |
## Verified models
|
|
@@ -44,16 +45,24 @@ Browse by domain: [Countdown](#domain-countdown) · [Graph Coloring](#domain-gra
|
|
| 44 |
<a id="domain-countdown"></a>
|
| 45 |
### Countdown
|
| 46 |
|
| 47 |
-
**
|
| 48 |
|
| 49 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 50 |
|
| 51 |
<a id="domain-graph_coloring"></a>
|
| 52 |
### Graph Coloring
|
| 53 |
|
| 54 |
-
**
|
| 55 |
|
| 56 |
-
|
|
|
|
|
|
|
|
|
|
| 57 |
|
| 58 |
<a id="domain-python_factors"></a>
|
| 59 |
### Python Factors
|
|
@@ -86,7 +95,7 @@ Transfer in progress; this model family has no verified exports in the index yet
|
|
| 86 |
<a id="model-falcon1b"></a>
|
| 87 |
### Falcon3-1B-Instruct
|
| 88 |
|
| 89 |
-
|
| 90 |
|
| 91 |
<a id="comparators"></a>
|
| 92 |
## Comparators and analysis scope
|
|
|
|
| 18 |
|
| 19 |
| Registered new training cells | Scientific records selected | Deployable exports selected | Verified exports |
|
| 20 |
| ---: | ---: | ---: | ---: |
|
| 21 |
+
| 75 | 49 | 49 | 6 |
|
| 22 |
|
| 23 |
Scientific records preserve the admitted result identity. Deployable exports additionally have available model weights; verified exports have completed remote verification. Cells outside the fixed selection are not silently counted, and unverified model folders are not linked below.
|
| 24 |
|
| 25 |
| Model | Verified / selected | Seeds in selection |
|
| 26 |
| --- | ---: | --- |
|
| 27 |
| [Qwen2.5-0.5B-Instruct](#model-qwen05b) | 0 / 25 | 43, 44, 45, 46, 47 |
|
| 28 |
+
| [Falcon3-1B-Instruct](#model-falcon1b) | 6 / 24 | 55, 56, 57, 58, 59 |
|
| 29 |
|
| 30 |
## Methods and collections
|
| 31 |
|
|
|
|
| 34 |
|
| 35 |
ReplayDr.GRPO augmented with semantic regularization and an isolated verified proposal sampler; a bundled comparison.
|
| 36 |
|
| 37 |
+
- [Falcon3-1B-Instruct · Countdown](https://huggingface.co/od2961/maxent-grpo-models/tree/main/experiments/E112-R1/Falcon3-1B-Instruct/countdown/verified_support_discovery) — 4 verified exports. Exact immutable revisions are in the model tables below.
|
| 38 |
+
- [Falcon3-1B-Instruct · Graph Coloring](https://huggingface.co/od2961/maxent-grpo-models/tree/main/experiments/E112-R1/Falcon3-1B-Instruct/graph_coloring/verified_support_discovery) — 2 verified exports. Exact immutable revisions are in the model tables below.
|
| 39 |
|
| 40 |
<a id="models"></a>
|
| 41 |
## Verified models
|
|
|
|
| 45 |
<a id="domain-countdown"></a>
|
| 46 |
### Countdown
|
| 47 |
|
| 48 |
+
**4 / 9** selected exports verified.
|
| 49 |
|
| 50 |
+
| Model | Method | Seed | Terminal export | Immutable revision |
|
| 51 |
+
| --- | --- | ---: | --- | --- |
|
| 52 |
+
| Falcon3-1B-Instruct | Verified-support semantic + proposal bundle | 55 | [step 3073](https://huggingface.co/od2961/maxent-grpo-models/tree/84e784c10b904062a75bdfd9c854b96a6a79df4d/experiments/E112-R1/Falcon3-1B-Instruct/countdown/verified_support_discovery/seed-55/step-03073) | [`84e784c10b90`](https://huggingface.co/od2961/maxent-grpo-models/tree/84e784c10b904062a75bdfd9c854b96a6a79df4d/experiments/E112-R1/Falcon3-1B-Instruct/countdown/verified_support_discovery/seed-55/step-03073) |
|
| 53 |
+
| Falcon3-1B-Instruct | Verified-support semantic + proposal bundle | 56 | [step 3073](https://huggingface.co/od2961/maxent-grpo-models/tree/240d248dbbb6cfa8b5dbd989de77e9f9b71b63a3/experiments/E112-R1/Falcon3-1B-Instruct/countdown/verified_support_discovery/seed-56/step-03073) | [`240d248dbbb6`](https://huggingface.co/od2961/maxent-grpo-models/tree/240d248dbbb6cfa8b5dbd989de77e9f9b71b63a3/experiments/E112-R1/Falcon3-1B-Instruct/countdown/verified_support_discovery/seed-56/step-03073) |
|
| 54 |
+
| Falcon3-1B-Instruct | Verified-support semantic + proposal bundle | 57 | [step 3073](https://huggingface.co/od2961/maxent-grpo-models/tree/60c7d9f383912e1969f5ecf5a38e15e998fe71a2/experiments/E112-R1/Falcon3-1B-Instruct/countdown/verified_support_discovery/seed-57/step-03073) | [`60c7d9f38391`](https://huggingface.co/od2961/maxent-grpo-models/tree/60c7d9f383912e1969f5ecf5a38e15e998fe71a2/experiments/E112-R1/Falcon3-1B-Instruct/countdown/verified_support_discovery/seed-57/step-03073) |
|
| 55 |
+
| Falcon3-1B-Instruct | Verified-support semantic + proposal bundle | 58 | [step 3073](https://huggingface.co/od2961/maxent-grpo-models/tree/27c3efc0ab3bf56284c60fd3965edf5c0a3e23a9/experiments/E112-R1/Falcon3-1B-Instruct/countdown/verified_support_discovery/seed-58/step-03073) | [`27c3efc0ab3b`](https://huggingface.co/od2961/maxent-grpo-models/tree/27c3efc0ab3bf56284c60fd3965edf5c0a3e23a9/experiments/E112-R1/Falcon3-1B-Instruct/countdown/verified_support_discovery/seed-58/step-03073) |
|
| 56 |
|
| 57 |
<a id="domain-graph_coloring"></a>
|
| 58 |
### Graph Coloring
|
| 59 |
|
| 60 |
+
**2 / 10** selected exports verified.
|
| 61 |
|
| 62 |
+
| Model | Method | Seed | Terminal export | Immutable revision |
|
| 63 |
+
| --- | --- | ---: | --- | --- |
|
| 64 |
+
| Falcon3-1B-Instruct | Verified-support semantic + proposal bundle | 55 | [step 3073](https://huggingface.co/od2961/maxent-grpo-models/tree/049be19138fb25ceb3f2178ad89c48937f333353/experiments/E112-R1/Falcon3-1B-Instruct/graph_coloring/verified_support_discovery/seed-55/step-03073) | [`049be19138fb`](https://huggingface.co/od2961/maxent-grpo-models/tree/049be19138fb25ceb3f2178ad89c48937f333353/experiments/E112-R1/Falcon3-1B-Instruct/graph_coloring/verified_support_discovery/seed-55/step-03073) |
|
| 65 |
+
| Falcon3-1B-Instruct | Verified-support semantic + proposal bundle | 56 | [step 3073](https://huggingface.co/od2961/maxent-grpo-models/tree/15e2f9cc9b4822cf6120dcd61500e7bb7a634a89/experiments/E112-R1/Falcon3-1B-Instruct/graph_coloring/verified_support_discovery/seed-56/step-03073) | [`15e2f9cc9b48`](https://huggingface.co/od2961/maxent-grpo-models/tree/15e2f9cc9b4822cf6120dcd61500e7bb7a634a89/experiments/E112-R1/Falcon3-1B-Instruct/graph_coloring/verified_support_discovery/seed-56/step-03073) |
|
| 66 |
|
| 67 |
<a id="domain-python_factors"></a>
|
| 68 |
### Python Factors
|
|
|
|
| 95 |
<a id="model-falcon1b"></a>
|
| 96 |
### Falcon3-1B-Instruct
|
| 97 |
|
| 98 |
+
[Browse verified collections](https://huggingface.co/od2961/maxent-grpo-models/tree/main/experiments/E112-R1/Falcon3-1B-Instruct). Domain tables above provide exact model revisions.
|
| 99 |
|
| 100 |
<a id="comparators"></a>
|
| 101 |
## Comparators and analysis scope
|