alt.freeweights.text / a3-rl-laion_exp_rpt_ghactions-v3-20-8B

From: laion <huggingface.co/laion>Newsgroups: alt.freeweights.textSubject:

a3-rl-laion_exp_rpt_ghactions-v3-20-8B

Date: 2026-11-15Commit: 87a2878Signed: 2DC28CC7B26A283B ✓

Model with 8.2B parameters. Safetensors weights, 16.7 GB in all. Published by laion under Apache-2.0.

8.2B params16.7GBa3-rl-laion_exp_rpt_ghactions-v3-20-8Bsafetensorspermissive

FactsParams8.2BSize16.7 GBFormatssafetensorsTasknot taggedLicenseApache-2.0From sourcepermissive

$ fw get laion/a3-rl-laion_exp_rpt_ghactions-v3-20-8B
# manifest sha256 e85cf72d631a… · 16.7 GB · 60 files
$ fw verify ./a3-rl-laion_exp_rpt_ghactions-v3-20-8B laion/a3-rl-laion_exp_rpt_ghactions-v3-20-8B

From sourceHosted (signed)laion/a3-rl-laion_exp_rpt_ghactions-v3-20-8BLast check: 2026-11-15 04:21 UTC

Run at home Get help

Get help: custom fine-tune or private deploy. We review your setup before we quote.

Fine-tune of GLM-4_7-swesmith-sandboxes-with_tests-oracle_verified_120s-maxeps-131k-fixthink.

Works with

Run8 Fine-tune7 Convert8 Scan2

Runs with

Ranked by fit to this model (safetensors), then by GitHub stars.

Transformers LocalAI tinygrad Triton Inference Server BentoML CTranslate2 ExLlamaV2 LitServe

Overview

Size: 16.7 GB
Weights: 16.4 GB (safetensors)
Parameters: 8.2B
Formats: safetensors
Task: not tagged
License: Apache-2.0 (permissive)
Files: 60, each with a sha256
Publisher: laion
Added: 2026-11-15
Section: alt.freeweights.text

Lineage

Original release from the publisher.

Base model: laion/GLM-4_7-swesmith-sandboxes-with_tests-oracle_verified_120s-maxeps-131k-fixthink

Where these bytes come from

Publisher at the pinned commit→You

Sources are tried in order. A file counts only when its size and sha256 match the manifest. Sources are checked when you ask for them, not on a timer.

Sources

Badge	Host	Last check	Result
From source	huggingface.co	2026-11-15 04:21 UTC	passed

Model card

From the publisher at commit 87a2878. Shown as plain text with links; images open at the publisher.

a3-rl-laion_exp_rpt_ghactions-v3 (step 20, 8B)#

RL (GRPO / rloo_n, SkyRL) fine-tune of laion/GLM-4_7-swesmith-sandboxes-with_tests-oracle_verified_120s-maxeps-131k-fixthink (Qwen3-8B architecture) on the agentic task set laion/exp_rpt_ghactions-v3.

This checkpoint is global_step_20, selected by the trailing-5-step EMA of reward/avg_raw_reward (EMA = 0.6933, raw reward at step 20 = 0.826), which is the sustained-best region of the run.

⚠️ Run status: COLLAPSED / NOT-CONVERGED#

This run collapsed after step 20 and did NOT exhaust to a healthy final state. The reward trajectory rose to a strong plateau (~0.73–0.83, pass@8 ≈ 1.0) around steps 16–23, then collapsed sharply:

step	avg_raw_reward	pass@8	policy_entropy
9	0.693	0.984	0.0844
16	0.777	1.000	0.0777
18	0.824	1.000	0.0830
20	0.826 (best)	1.000	0.0712
23	0.717	0.984	0.0568
25	0.430	0.938	0.0304
26	0.107	0.375	0.0070
27	0.002 (final)	0.016	0.0001

Entropy decayed toward zero alongside the reward crash — the classic entropy-collapse signature. The FINAL training step (27) was reward ≈ 0.002 / near-zero entropy (fully collapsed). This -20-8B checkpoint is the pre-collapse peak, NOT the final state of training.

Other pre-collapse checkpoints also exist in the experiment exports (global_step_15 was a transient dip to 0.0; global_step_25 was already collapsing at reward 0.43). Step 20 is the highest-EMA aligned checkpoint and is the one published here.

Treat this as a low-quality / non-converged research artifact, not a clean production model.

Training config#

Algorithm: GRPO (rloo_n advantage), SkyRL fully-async trainer, FSDP2
Base: laion/GLM-4_7-swesmith-sandboxes-with_tests-oracle_verified_120s-maxeps-131k-fixthink
Dataset: laion/exp_rpt_ghactions-v3
hf_save_interval = 5, max_steps = 80 (run collapsed/exhausted at step ~27)
14 nodes × 4 GH200 (Jupiter), TP=1

See rl_config.json and training_logs/ (metrics CSVs, reward plot, per-trial results from parse_skyrl_metrics.py) in this repo.

Training Traces#

Training-time Daytona/Harbor rollouts for this run are uploaded as a companion dataset: open-athena/a3-rl-laion_exp_rpt_ghactions-v3

The dataset contains the last episode of each trial (per make_and_upload_trace_dataset --episodes last) — the same rollouts the policy was trained on after rollback / truncation.

Files

60 files · 16.7 GBWeights 16.4 GB (safetensors). Sizes and sha256 from the signed manifest.

Show all 60 files with their sha256

File	Size	sha256
.gitattributes	2.41 KB	d00212299d26d6c70ac671aed4a1084af61d7d8b986d7dbd93c1e4b8b3c7d1c8
README.md	2.94 KB	2e6ae42dc03a20948c80c4e612d270c895ad49ed32e53797696b96ff43d69802
added_tokens.json	707 B	c0284b582e14987fbd3d5a2cb2bd139084371ed9acbae488829a1c900833c680
chat_template.jinja	4.17 KB	a55ee1b1660128b7098723e0abcd92caa0788061051c62d51cbe87d9cf1974d8
config.json	1.54 KB	ed36641e183b3d95dcefad8906aa22d183c445e4b582c8a0ca7d71a713c85f2d
generation_config.json	188 B	3b101cd89b140b757d0bca469e75570e75789a2423ada90a33ec17707dfacf08
merges.txt	1.67 MB	8831e4f1a044471340f7c0a83d7bd71306a5b867e95fd870f74d0c5308a904d5
model-00001-of-00004.safetensors	4.90 GB	1ec36e6dda765ce902caa98d1bfe98cfb17d755bd6123ba3cc291553fd522227
model-00002-of-00004.safetensors	4.92 GB	9597d4c821ec2d310afdf720381605b4b337ac581a803993b35579d3b04905a4
model-00003-of-00004.safetensors	4.98 GB	97955e156ba2fc33dd1923578d4a2ef62bd2e16654fffe94e927e097298f5ed8
model-00004-of-00004.safetensors	1.58 GB	6857d5b349babfcacc752f076b78681060cc7fd9c7341fbf3d1acfb2f862a793
model.safetensors.index.json	32.9 KB	e7e4bdc58d3302c97357f27979b270987dd620cf0cd9b1c130e3a51c9d64df95
rl_config.json	8.22 KB	b5dc9c0b55c74e4f26ae67d8a76e5053bedefb018ce317ee67f976bbac7dacf4
special_tokens_map.json	613 B	76862e765266b85aa9459767e33cbaf13970f327a0e88d1c65846c2ddd3a1ecd
tokenizer.json	11.4 MB	aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4
tokenizer_config.json	5.43 KB	3e27b302e9844c5b9648a8dbc1958fbbf3bda640e07302f53c17ac7a3192def5
training_logs/20260602_114527_metrics_job_528089.csv	3.17 KB	e9f4eac794e2391772a27653732fd8341a7774dad512352cdb2cc4cab3ebeffc
training_logs/20260602_114527_metrics_job_537282.csv	3.72 KB	7abc744acf11ec9f5b98a893e4728fbf336e7eb926ca2e589bf435829add950d
training_logs/20260602_114527_metrics_job_537283.csv	3.79 KB	20f9b672df7d1646c7bc2855185464b71ef3f61f7d160cf4cd77d281729d164f
training_logs/20260602_114527_metrics_job_537284.csv	3.16 KB	93fdde8e84d2358604a83cc4c61fd9ddd1f13a2b0b6735a4420369c6f1111b20
training_logs/20260602_114527_metrics_job_537285.csv	5.50 KB	8d9404a171cb34ce6cdb4eba650f78d0a90cb142ad4fffde77edf11e4a4c1597
training_logs/20260602_114527_metrics_job_537286.csv	3.29 KB	54e45137faebc274b2cb997a36e560ab22f7378debcb4258e6da54e84203c4d2
training_logs/20260602_114527_metrics_job_537288.csv	7.18 KB	2c9402d86ce98ca4ed40a9d1b0a85829520c5a14e87ba535f74fba9051bc2899
training_logs/20260602_114527_metrics_report.md	25.9 KB	0ab392731b9d4a6376a79c97523bb8b1b5395e14567e2d4e6c73c9e2b0339467
training_logs/20260602_114527_metrics_table.csv	17.7 KB	37b4044a2f7b63d242da87a98499184ad89bb6426a520ae6f69a23705766ad4d
training_logs/20260602_114527_reward_vs_steps.png	216 KB	a2425917ae5c4214b53f063296b999443051a829497700605f1838c7b7f71022
training_logs/20260602_114527_trial_results.csv	1.75 MB	52152eb2731a644577de27f05a515789c1614ddb10f3070a58405b6a146b7814
training_logs/20260602_114527_turn_count_distribution.png	126 KB	ffdea374eeb417a5ab2ee0207ddf1c0656b9cf30905345c54e425f48eb291693
training_logs/20260602_114527_vllm_metrics_job_528087.csv	56.0 KB	549b292ec3bd35f5683e57d569563eb78dfcbb1366ec3792184ffcbdfadd2ec6
training_logs/20260602_114527_vllm_metrics_job_528088.csv	61.1 KB	efe13aaa377171d2b3bbdcb18de98458db35044dcdb530f2ca74a0a16a0afe98
training_logs/20260602_114527_vllm_metrics_job_528089.csv	98.4 KB	1dd239d7361a950b604ee3d141f917d5eba6c41c25160e667285cff631abf242
training_logs/20260602_114527_vllm_metrics_job_528090.csv	41.9 KB	8272d8fa9945b9e0f3f9aba41fa6f6f55a36e390859a3e89ada305e3fe2e13ef
training_logs/20260602_114527_vllm_metrics_job_528092.csv	31.8 KB	b55c78f11a77e696b219da12734bbf5853c240ca2d11ddfd8539d48113fa6f3e
training_logs/20260602_114527_vllm_metrics_job_528093.csv	76.5 KB	e864fd2e19a70a43908366830c768320c406ded8ba17193d0cd48eae5e92b676
training_logs/20260602_114527_vllm_metrics_job_536586.csv	38.2 KB	48a97168b39ab939a4c17e813e5a9274889d6774d5f54cac2b8fa2a7fdad561d
training_logs/20260602_114527_vllm_metrics_job_536587.csv	13.1 KB	5e82d3a16a5d9f663aecb472a90a344fe1a0b12c360688d7088c5032c03a4b3c
training_logs/20260602_114527_vllm_metrics_job_537282.csv	178 KB	70699e2e7314b0121cac54972c036bee0b2b7bd388b89d8eccb109e126098957
training_logs/20260602_114527_vllm_metrics_job_537283.csv	115 KB	0cee7ddeee37147a8076c9e8ade2c2006a4fb0b6b0f433b1c59fb63a590622d2
training_logs/20260602_114527_vllm_metrics_job_537284.csv	123 KB	7ba7a34413aac9453828d83d4671d2c1f02e98642070f1778cf49ceb609c750b
training_logs/20260602_114527_vllm_metrics_job_537285.csv	133 KB	4d797691dfae7199e5df8682d6712e86f5adcb06b68f68d9551370f25d11ca8e
training_logs/20260602_114527_vllm_metrics_job_537286.csv	142 KB	60551b6840a4fb4532a9316284be0550bac078aec3ac6e85f94ddb70d40b27ab
training_logs/20260602_114527_vllm_metrics_job_537288.csv	257 KB	043982ae7ff1fae1e566257b54b2b4b665e637c1d8c19d8788da348b13ecd4a2
training_logs/20260602_114527_vllm_metrics_table.csv	1.56 MB	bfdc981901d5ecc6c81138c5f7afe52273aa34f52a1a83ed9be17270188ce924
training_logs/a3-rl-laion_exp_rpt_ghactions-v3_528087.out	7.23 MB	d1ee797cf0c288bb2240ec3974e37eb47f4f7ecc646e5a2e3666902b52bea035
training_logs/a3-rl-laion_exp_rpt_ghactions-v3_528088.out	7.97 MB	1bfdf23dfbd732dbf0c8326db8ac7c8b587b3d770a86296d89522a8b7caad3e1
training_logs/a3-rl-laion_exp_rpt_ghactions-v3_528089.out	28.5 MB	a96871686df808d5a198d262f89d30070d875bf7dd990ecb64bac052e78dfc39
training_logs/a3-rl-laion_exp_rpt_ghactions-v3_528090.out	5.70 MB	6294aaa79a5f078754a6c65b51f2dd1a140371e9d286e40f6237111616bb701c
training_logs/a3-rl-laion_exp_rpt_ghactions-v3_528091.out	277 KB	f9dafd7e65f8266ad29fb5960479c162cd28f39c9e9b66d74820ec35c42abb16
training_logs/a3-rl-laion_exp_rpt_ghactions-v3_528092.out	4.90 MB	dda27ed421d77ae28bd24f7dd7514f268ab2a9d01c8e2e8866a941a371674d8a
training_logs/a3-rl-laion_exp_rpt_ghactions-v3_528093.out	8.38 MB	9ff36f3b7b0c3f9e738fb0d84e96acdd0e40a076eaa4bfd35fbfa591d6537051
training_logs/a3-rl-laion_exp_rpt_ghactions-v3_536586.out	4.84 MB	beba8fcc6a8dd7553febac75a7724089df017998431d49c85f59616d6bc16e1b
training_logs/a3-rl-laion_exp_rpt_ghactions-v3_536587.out	2.68 MB	8f1c660b19905ee71a76dcac6b975b3f16b75cca9cc198adf1abd38ce504ccff
training_logs/a3-rl-laion_exp_rpt_ghactions-v3_537282.out	27.8 MB	4ee4d6b56303cced1ee2803952522c86ed83395372dddc2ea1cc4bff0a449803
training_logs/a3-rl-laion_exp_rpt_ghactions-v3_537283.out	28.1 MB	e3f16ad74e3d31e0ae14fb3b88b8a2ffd9f709709929e08f1f8a601e127a7a65
training_logs/a3-rl-laion_exp_rpt_ghactions-v3_537284.out	23.8 MB	a1845f86602ebd7991ae42331cd627bff046f421694bc39d8a6beae9e80fb307
training_logs/a3-rl-laion_exp_rpt_ghactions-v3_537285.out	51.9 MB	ce9236530e2e8e67b3f5fcf6fd35461737c4538e7bfd2bd2d761431cca95d2ff
training_logs/a3-rl-laion_exp_rpt_ghactions-v3_537286.out	26.2 MB	5c12ac9cb25ab0065faff28442ca7af381576018ed8b00c3585a33b7dd4fc15d
training_logs/a3-rl-laion_exp_rpt_ghactions-v3_537287.out	51.3 KB	d0bee110c0ee1fe557e6dce2430ded566c305417add7c91d6630c92960ca7b4a
training_logs/a3-rl-laion_exp_rpt_ghactions-v3_537288.out	49.7 MB	32b44764d2d79bc38b18aeac2a2dc75c2411662831612be81f0a7bb00ffa2f2f
vocab.json	2.78 MB	ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910

VersionsEach time the pinned upstream commit or file list changed in this repo. Newest first.

Upstream commit	Date	Contents
87a28781d0 pinned now	2026-11-15 04:21 UTC	60 files · 16.7 GB

Use it

$ fw get laion/a3-rl-laion_exp_rpt_ghactions-v3-20-8B
# manifest sha256 e85cf72d631a… · 16.7 GB · 60 files
$ fw verify ./a3-rl-laion_exp_rpt_ghactions-v3-20-8B laion/a3-rl-laion_exp_rpt_ghactions-v3-20-8B

fw.pyChecks the signature and every sha256, tries each source in order, resumes.

$ python3 fw.py get https://freeweights.allternit.com/manifests/laion/a3-rl-laion_exp_rpt_ghactions-v3-20-8B.json --dir ./a3-rl-laion_exp_rpt_ghactions-v3-20-8B
$ python3 fw.py verify ./a3-rl-laion_exp_rpt_ghactions-v3-20-8B ./laion__a3-rl-laion_exp_rpt_ghactions-v3-20-8B.json

hfThe Hugging Face CLI with one setting. Same repo id.

$ HF_ENDPOINT=https://freeweights.allternit.com hf download laion/a3-rl-laion_exp_rpt_ghactions-v3-20-8B --revision 87a28781d01cce4b44a9ec542356b5757c8fad28 --local-dir ./a3-rl-laion_exp_rpt_ghactions-v3-20-8B

Run at home

License

Name: Apache License 2.0
Id: Apache-2.0
Class: Permissive
Redistribution: yes
Listing: Hosted (signed)
Decided: License as declared by the publisher. No person read it. 2026-11-15
Text in: README.md
Text sha256: 2e6ae42dc03a20948c80c4e612d270c895ad49ed32e53797696b96ff43d69802
Terms gate: No extra terms beyond the license.

Read the license text

Licenses are checked against the license table. A custom license passes only after a person reads it.

Provenance

From: laion <huggingface.co/laion>Newsgroups: alt.freeweights.textSubject: a3-rl-laion_exp_rpt_ghactions-v3-20-8BDate: 2026-11-15Commit: 87a2878Signed: 2DC28CC7B26A283B ✓

Upstream: laion/a3-rl-laion_exp_rpt_ghactions-v3-20-8B
Pinned commit: 87a28781d01cce4b44a9ec542356b5757c8fad28
Manifest: a3-rl-laion_exp_rpt_ghactions-v3-20-8B.json · .minisig
Manifest sha256: e85cf72d631a1f561459725177b84008db4940a99ff114dc961046f6e304886b
Signature: minisign, Ed25519, key 2DC28CC7B26A283B
Created: 2026-11-15 04:21 UTC
Generator: freeweights-manifest/0.1.0
Takedowns: None.

How to verify

$ curl -O https://freeweights.allternit.com/manifests/laion/a3-rl-laion_exp_rpt_ghactions-v3-20-8B.json
$ curl -O https://freeweights.allternit.com/manifests/laion/a3-rl-laion_exp_rpt_ghactions-v3-20-8B.json.minisig
$ minisign -Vm a3-rl-laion_exp_rpt_ghactions-v3-20-8B.json -P RWQ7KGqyx4zCLXtUFOqvfv8L3g+8fpKJ1FsuxPyOYMF+4KufbUxKq1Oq
$ shasum -a 256 <file>   # compare with the Files tab

A manifest whose signature fails must be rejected. A file whose size or sha256 does not match must be discarded. Verify a download

Model tree

Parent

laion/GLM-4_7-swesmith-sandboxes-with_tests-oracle_verified_120s-maxeps-131k-fixthink · this model is a finetune of it

No model in the catalog names this one as its parent.

Get it

Download

$ HF_ENDPOINT=https://freeweights.allternit.com hf download laion/a3-rl-laion_exp_rpt_ghactions-v3-20-8B

Open in Freeweights Studio

Pull, verify and run it in the Allternit app on Desktop, web or phone.

Signed manifest

minisign, key 2DC28CC7B26A283B. a3-rl-laion_exp_rpt_ghactions-v3-20-8B.json. Provenance. Takedown history: none.

Source

huggingface.co/laion/a3-rl-laion_exp_rpt_ghactions-v3-20-8B

Related models

All of alt.freeweights.text →

From: laion <huggingface.co/laion>Subject: a2-rl-e2egit_large-50-32B

From: laion <huggingface.co/laion>Subject: a3-rl-DCAgent_exp_rpt_e2egit-large-15-8B

From: laion <huggingface.co/laion>Subject: a3-rl-DCAgent_exp_rpt_e2egit-v2-10-8B

From: laion <huggingface.co/laion>Subject: a3-rl-DCAgent_exp_rpt_pymethods2test-large-80-8B

From: laion <huggingface.co/laion>Subject: a3-rl-DCAgent_exp_rpt_unitsyn-python-large-10-8B

From: laion <huggingface.co/laion>Subject: a3-rl-DCAgent_mix_h2_language_proportional-65-8B