Hugging Face's logo

Datasets:
arvindcr4
/
tinker-rl-bench-checkpoints

Tasks:
Text Generation
Reinforcement Learning
Modalities:
Tabular
Text
Formats:
json
Size:
< 1K
Tags:
tinker
grpo
ppo
rlhf
checkpoints
manifest
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
arvindcr4 commited on
Commit
6df149a
·
verified ·
1 Parent(s): 22bd9a5

Initial: 28 Tinker training runs, 66 checkpoint URIs + enriched metadata

Browse files
Files changed (8) hide show
  1. README.md +97 -0
  2. all_results_consolidated.json +1882 -0
  3. arithmetic_checkpoints.jsonl +6 -0
  4. checkpoints.jsonl +66 -0
  5. distillation_off_checkpoints.jsonl +14 -0
  6. heldout_gsm8k.json +0 -0
  7. master_results.json +4183 -0
  8. training_runs.jsonl +28 -0
README.md ADDED
@@ -0,0 +1,97 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ task_categories:
4
+ - text-generation
5
+ - reinforcement-learning
6
+ tags:
7
+ - tinker
8
+ - grpo
9
+ - ppo
10
+ - rlhf
11
+ - checkpoints
12
+ - manifest
13
+ pretty_name: TinkerRL-Bench Checkpoint Manifest
14
+ size_categories:
15
+ - n<1K
16
+ configs:
17
+ - config_name: training_runs
18
+ data_files: training_runs.jsonl
19
+ - config_name: checkpoints
20
+ data_files: checkpoints.jsonl
21
+ ---
22
+
23
+ # TinkerRL-Bench Checkpoint Manifest
24
+
25
+ A catalogue of every Tinker training run and checkpoint referenced by our
26
+ NeurIPS paper *"A Unified Benchmark for RL Post-Training of Language
27
+ Models"* ([repo](https://github.com/pes-llm-research/tinker-rl-lab)).
28
+
29
+ Because Tinker stores weights behind an authenticated `tinker://...` URI
30
+ (only the account that ran the training can materialise them), this
31
+ dataset does **not** contain the raw `.safetensors`/archive blobs — it
32
+ contains the canonical pointer table and full training metadata so
33
+ anyone with a Tinker API key can fetch the exact artifact that produced
34
+ a given result.
35
+
36
+ ## Contents
37
+
38
+ | File | Rows | Description |
39
+ |------|------|-------------|
40
+ | `training_runs.jsonl` | 28 | One row per training run: `training_run_id`, `experiment`, `model`, `model_short`, `task`, `seed`, `rank`, `lr`, `group_size`, `steps`, `platform`, `last10_avg`, `peak_accuracy`, `last10_accuracy`, `weight_checkpoints[]`, `sampler_weight_checkpoints[]`. |
41
+ | `checkpoints.jsonl` | 66 | One row per `(training_run_id, kind, step)`: `tinker_uri`, `kind ∈ {weights, sampler_weights}`, `step`, `is_final`, full joined metadata, source files that reference it. |
42
+ | `arithmetic_checkpoints.jsonl`, `distillation_off_checkpoints.jsonl` | — | Original per-step checkpoint indices committed to the repo. |
43
+ | `heldout_gsm8k.json` | — | Held-out GSM8K evaluation of the top-10 Tinker checkpoints. |
44
+ | `all_results_consolidated.json`, `master_results.json` | — | Source of truth for run-level metadata. |
45
+
46
+ ## Coverage
47
+
48
+ - **28 unique Tinker training runs** across `tinker-rl-lab-world-class`,
49
+ `tinker-structural-ceiling`, `tinker-rl-scaling`, and `skyrl-tinker`
50
+ W&B projects.
51
+ - **66 distinct checkpoint URIs**: 20 intermediate `weights/` (arithmetic
52
+ + distillation-off sweeps with per-20 or per-50 step snapshots) and 45
53
+ `sampler_weights/` (final + per-step samplers). One run
54
+ (`38d13280...`) has 10 mid-training weight snapshots at steps
55
+ 50–500.
56
+ - Models covered include: Llama-3.1-8B-Base/Instruct, Llama-3.2-{1B,3B},
57
+ Qwen3-{0.6B, 1.7B, 4B, 8B-Base/Instruct, 14B, 30B-MoE, 32B, 235B},
58
+ Qwen3.5-{4B, 27B}, Gemma-2-{2B, 9B}, Nemotron-120B, DeepSeek-V3.1,
59
+ GPT-OSS-20B, Kimi-K2.
60
+
61
+ ## Materialising weights
62
+
63
+ Tinker weights are downloadable by the training account with:
64
+
65
+ ```python
66
+ import os, tinker, urllib.request
67
+ sc = tinker.ServiceClient() # reads TINKER_API_KEY
68
+ rc = sc.create_rest_client()
69
+ fut = rc.get_checkpoint_archive_url_from_tinker_path(
70
+ "tinker://<training_run_id>:train:0/sampler_weights/final"
71
+ )
72
+ url = fut.result().url
73
+ urllib.request.urlretrieve(url, "archive.tar")
74
+ ```
75
+
76
+ The returned archive contains LoRA adapters + tokenizer and can be
77
+ loaded into the corresponding base model with `tinker_cookbook`.
78
+
79
+ ## Companion datasets
80
+
81
+ - [`arvindcr4/tinker-rl-bench-wandb`](https://huggingface.co/datasets/arvindcr4/tinker-rl-bench-wandb)
82
+ — 334 W&B runs + 9,255 history rows from the same experiments.
83
+
84
+ ## Citation
85
+
86
+ ```bibtex
87
+ @misc{tinkerrlbench2026,
88
+ title = {A Unified Benchmark for RL Post-Training of Language Models},
89
+ author = {Arvind, C. R. and Jeyaraj, Sandhya},
90
+ year = {2026},
91
+ note = {NeurIPS submission, https://github.com/pes-llm-research/tinker-rl-lab}
92
+ }
93
+ ```
94
+
95
+ ## License
96
+
97
+ Apache 2.0.
all_results_consolidated.json ADDED
@@ -0,0 +1,1882 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [
2
+ {
3
+ "experiment": "cross_tool_llama-8b-inst",
4
+ "model": "meta-llama/Llama-3.1-8B-Instruct",
5
+ "model_short": "llama-8b-inst",
6
+ "task": "tool_use",
7
+ "seed": 42,
8
+ "rank": 32,
9
+ "lr": 3e-05,
10
+ "group": 8,
11
+ "steps": 30,
12
+ "run_id": "ca2e3a24-7401-5770-af34-a0d27177aeaa:train:0",
13
+ "checkpoint": "tinker://ca2e3a24-7401-5770-af34-a0d27177aeaa:train:0/sampler_weights/final",
14
+ "first5_avg": 0.0,
15
+ "last10_avg": 0.0,
16
+ "peak": 0.0,
17
+ "zero_loss_pct": 100.0,
18
+ "zero_reward_pct": 100.0,
19
+ "reward_trace": [
20
+ 0.0,
21
+ 0.0,
22
+ 0.0,
23
+ 0.0,
24
+ 0.0,
25
+ 0.0,
26
+ 0.0,
27
+ 0.0,
28
+ 0.0,
29
+ 0.0,
30
+ 0.0,
31
+ 0.0,
32
+ 0.0,
33
+ 0.0,
34
+ 0.0,
35
+ 0.0,
36
+ 0.0,
37
+ 0.0,
38
+ 0.0,
39
+ 0.0,
40
+ 0.0,
41
+ 0.0,
42
+ 0.0,
43
+ 0.0,
44
+ 0.0,
45
+ 0.0,
46
+ 0.0,
47
+ 0.0,
48
+ 0.0,
49
+ 0.0
50
+ ],
51
+ "step_log": [
52
+ {
53
+ "step": 1,
54
+ "loss": 0.0,
55
+ "reward": 0.0,
56
+ "zvf": 1.0,
57
+ "gu": 0.0
58
+ },
59
+ {
60
+ "step": 2,
61
+ "loss": 0.0,
62
+ "reward": 0.0,
63
+ "zvf": 1.0,
64
+ "gu": 0.0
65
+ },
66
+ {
67
+ "step": 3,
68
+ "loss": 0.0,
69
+ "reward": 0.0,
70
+ "zvf": 1.0,
71
+ "gu": 0.0
72
+ },
73
+ {
74
+ "step": 4,
75
+ "loss": 0.0,
76
+ "reward": 0.0,
77
+ "zvf": 1.0,
78
+ "gu": 0.0
79
+ },
80
+ {
81
+ "step": 5,
82
+ "loss": 0.0,
83
+ "reward": 0.0,
84
+ "zvf": 1.0,
85
+ "gu": 0.0
86
+ },
87
+ {
88
+ "step": 6,
89
+ "loss": 0.0,
90
+ "reward": 0.0,
91
+ "zvf": 1.0,
92
+ "gu": 0.0
93
+ },
94
+ {
95
+ "step": 7,
96
+ "loss": 0.0,
97
+ "reward": 0.0,
98
+ "zvf": 1.0,
99
+ "gu": 0.0
100
+ },
101
+ {
102
+ "step": 8,
103
+ "loss": 0.0,
104
+ "reward": 0.0,
105
+ "zvf": 1.0,
106
+ "gu": 0.0
107
+ },
108
+ {
109
+ "step": 9,
110
+ "loss": 0.0,
111
+ "reward": 0.0,
112
+ "zvf": 1.0,
113
+ "gu": 0.0
114
+ },
115
+ {
116
+ "step": 10,
117
+ "loss": 0.0,
118
+ "reward": 0.0,
119
+ "zvf": 1.0,
120
+ "gu": 0.0
121
+ },
122
+ {
123
+ "step": 11,
124
+ "loss": 0.0,
125
+ "reward": 0.0,
126
+ "zvf": 1.0,
127
+ "gu": 0.0
128
+ },
129
+ {
130
+ "step": 12,
131
+ "loss": 0.0,
132
+ "reward": 0.0,
133
+ "zvf": 1.0,
134
+ "gu": 0.0
135
+ },
136
+ {
137
+ "step": 13,
138
+ "loss": 0.0,
139
+ "reward": 0.0,
140
+ "zvf": 1.0,
141
+ "gu": 0.0
142
+ },
143
+ {
144
+ "step": 14,
145
+ "loss": 0.0,
146
+ "reward": 0.0,
147
+ "zvf": 1.0,
148
+ "gu": 0.0
149
+ },
150
+ {
151
+ "step": 15,
152
+ "loss": 0.0,
153
+ "reward": 0.0,
154
+ "zvf": 1.0,
155
+ "gu": 0.0
156
+ },
157
+ {
158
+ "step": 16,
159
+ "loss": 0.0,
160
+ "reward": 0.0,
161
+ "zvf": 1.0,
162
+ "gu": 0.0
163
+ },
164
+ {
165
+ "step": 17,
166
+ "loss": 0.0,
167
+ "reward": 0.0,
168
+ "zvf": 1.0,
169
+ "gu": 0.0
170
+ },
171
+ {
172
+ "step": 18,
173
+ "loss": 0.0,
174
+ "reward": 0.0,
175
+ "zvf": 1.0,
176
+ "gu": 0.0
177
+ },
178
+ {
179
+ "step": 19,
180
+ "loss": 0.0,
181
+ "reward": 0.0,
182
+ "zvf": 1.0,
183
+ "gu": 0.0
184
+ },
185
+ {
186
+ "step": 20,
187
+ "loss": 0.0,
188
+ "reward": 0.0,
189
+ "zvf": 1.0,
190
+ "gu": 0.0
191
+ },
192
+ {
193
+ "step": 21,
194
+ "loss": 0.0,
195
+ "reward": 0.0,
196
+ "zvf": 1.0,
197
+ "gu": 0.0
198
+ },
199
+ {
200
+ "step": 22,
201
+ "loss": 0.0,
202
+ "reward": 0.0,
203
+ "zvf": 1.0,
204
+ "gu": 0.0
205
+ },
206
+ {
207
+ "step": 23,
208
+ "loss": 0.0,
209
+ "reward": 0.0,
210
+ "zvf": 1.0,
211
+ "gu": 0.0
212
+ },
213
+ {
214
+ "step": 24,
215
+ "loss": 0.0,
216
+ "reward": 0.0,
217
+ "zvf": 1.0,
218
+ "gu": 0.0
219
+ },
220
+ {
221
+ "step": 25,
222
+ "loss": 0.0,
223
+ "reward": 0.0,
224
+ "zvf": 1.0,
225
+ "gu": 0.0
226
+ },
227
+ {
228
+ "step": 26,
229
+ "loss": 0.0,
230
+ "reward": 0.0,
231
+ "zvf": 1.0,
232
+ "gu": 0.0
233
+ },
234
+ {
235
+ "step": 27,
236
+ "loss": 0.0,
237
+ "reward": 0.0,
238
+ "zvf": 1.0,
239
+ "gu": 0.0
240
+ },
241
+ {
242
+ "step": 28,
243
+ "loss": 0.0,
244
+ "reward": 0.0,
245
+ "zvf": 1.0,
246
+ "gu": 0.0
247
+ },
248
+ {
249
+ "step": 29,
250
+ "loss": 0.0,
251
+ "reward": 0.0,
252
+ "zvf": 1.0,
253
+ "gu": 0.0
254
+ },
255
+ {
256
+ "step": 30,
257
+ "loss": 0.0,
258
+ "reward": 0.0,
259
+ "zvf": 1.0,
260
+ "gu": 0.0
261
+ }
262
+ ],
263
+ "timestamp": "2026-04-18T16:01:52.013977",
264
+ "platform": "tinker",
265
+ "peak_accuracy": 0.0,
266
+ "last10_accuracy": 0.0
267
+ },
268
+ {
269
+ "experiment": "cross_tool_qwen3-32b",
270
+ "model": "Qwen/Qwen3-32B",
271
+ "model_short": "qwen3-32b",
272
+ "task": "tool_use",
273
+ "seed": 42,
274
+ "rank": 32,
275
+ "lr": 3e-05,
276
+ "group": 8,
277
+ "steps": 30,
278
+ "run_id": "73ff186a-d8e3-50c3-afb8-2b863cd09579:train:0",
279
+ "checkpoint": "tinker://73ff186a-d8e3-50c3-afb8-2b863cd09579:train:0/sampler_weights/final",
280
+ "first5_avg": 0.0,
281
+ "last10_avg": 0.0,
282
+ "peak": 0.0,
283
+ "zero_loss_pct": 100.0,
284
+ "zero_reward_pct": 100.0,
285
+ "reward_trace": [
286
+ 0.0,
287
+ 0.0,
288
+ 0.0,
289
+ 0.0,
290
+ 0.0,
291
+ 0.0,
292
+ 0.0,
293
+ 0.0,
294
+ 0.0,
295
+ 0.0,
296
+ 0.0,
297
+ 0.0,
298
+ 0.0,
299
+ 0.0,
300
+ 0.0,
301
+ 0.0,
302
+ 0.0,
303
+ 0.0,
304
+ 0.0,
305
+ 0.0,
306
+ 0.0,
307
+ 0.0,
308
+ 0.0,
309
+ 0.0,
310
+ 0.0,
311
+ 0.0,
312
+ 0.0,
313
+ 0.0,
314
+ 0.0,
315
+ 0.0
316
+ ],
317
+ "step_log": [
318
+ {
319
+ "step": 1,
320
+ "loss": 0.0,
321
+ "reward": 0.0,
322
+ "zvf": 1.0,
323
+ "gu": 0.0
324
+ },
325
+ {
326
+ "step": 2,
327
+ "loss": 0.0,
328
+ "reward": 0.0,
329
+ "zvf": 1.0,
330
+ "gu": 0.0
331
+ },
332
+ {
333
+ "step": 3,
334
+ "loss": 0.0,
335
+ "reward": 0.0,
336
+ "zvf": 1.0,
337
+ "gu": 0.0
338
+ },
339
+ {
340
+ "step": 4,
341
+ "loss": 0.0,
342
+ "reward": 0.0,
343
+ "zvf": 1.0,
344
+ "gu": 0.0
345
+ },
346
+ {
347
+ "step": 5,
348
+ "loss": 0.0,
349
+ "reward": 0.0,
350
+ "zvf": 1.0,
351
+ "gu": 0.0
352
+ },
353
+ {
354
+ "step": 6,
355
+ "loss": 0.0,
356
+ "reward": 0.0,
357
+ "zvf": 1.0,
358
+ "gu": 0.0
359
+ },
360
+ {
361
+ "step": 7,
362
+ "loss": 0.0,
363
+ "reward": 0.0,
364
+ "zvf": 1.0,
365
+ "gu": 0.0
366
+ },
367
+ {
368
+ "step": 8,
369
+ "loss": 0.0,
370
+ "reward": 0.0,
371
+ "zvf": 1.0,
372
+ "gu": 0.0
373
+ },
374
+ {
375
+ "step": 9,
376
+ "loss": 0.0,
377
+ "reward": 0.0,
378
+ "zvf": 1.0,
379
+ "gu": 0.0
380
+ },
381
+ {
382
+ "step": 10,
383
+ "loss": 0.0,
384
+ "reward": 0.0,
385
+ "zvf": 1.0,
386
+ "gu": 0.0
387
+ },
388
+ {
389
+ "step": 11,
390
+ "loss": 0.0,
391
+ "reward": 0.0,
392
+ "zvf": 1.0,
393
+ "gu": 0.0
394
+ },
395
+ {
396
+ "step": 12,
397
+ "loss": 0.0,
398
+ "reward": 0.0,
399
+ "zvf": 1.0,
400
+ "gu": 0.0
401
+ },
402
+ {
403
+ "step": 13,
404
+ "loss": 0.0,
405
+ "reward": 0.0,
406
+ "zvf": 1.0,
407
+ "gu": 0.0
408
+ },
409
+ {
410
+ "step": 14,
411
+ "loss": 0.0,
412
+ "reward": 0.0,
413
+ "zvf": 1.0,
414
+ "gu": 0.0
415
+ },
416
+ {
417
+ "step": 15,
418
+ "loss": 0.0,
419
+ "reward": 0.0,
420
+ "zvf": 1.0,
421
+ "gu": 0.0
422
+ },
423
+ {
424
+ "step": 16,
425
+ "loss": 0.0,
426
+ "reward": 0.0,
427
+ "zvf": 1.0,
428
+ "gu": 0.0
429
+ },
430
+ {
431
+ "step": 17,
432
+ "loss": 0.0,
433
+ "reward": 0.0,
434
+ "zvf": 1.0,
435
+ "gu": 0.0
436
+ },
437
+ {
438
+ "step": 18,
439
+ "loss": 0.0,
440
+ "reward": 0.0,
441
+ "zvf": 1.0,
442
+ "gu": 0.0
443
+ },
444
+ {
445
+ "step": 19,
446
+ "loss": 0.0,
447
+ "reward": 0.0,
448
+ "zvf": 1.0,
449
+ "gu": 0.0
450
+ },
451
+ {
452
+ "step": 20,
453
+ "loss": 0.0,
454
+ "reward": 0.0,
455
+ "zvf": 1.0,
456
+ "gu": 0.0
457
+ },
458
+ {
459
+ "step": 21,
460
+ "loss": 0.0,
461
+ "reward": 0.0,
462
+ "zvf": 1.0,
463
+ "gu": 0.0
464
+ },
465
+ {
466
+ "step": 22,
467
+ "loss": 0.0,
468
+ "reward": 0.0,
469
+ "zvf": 1.0,
470
+ "gu": 0.0
471
+ },
472
+ {
473
+ "step": 23,
474
+ "loss": 0.0,
475
+ "reward": 0.0,
476
+ "zvf": 1.0,
477
+ "gu": 0.0
478
+ },
479
+ {
480
+ "step": 24,
481
+ "loss": 0.0,
482
+ "reward": 0.0,
483
+ "zvf": 1.0,
484
+ "gu": 0.0
485
+ },
486
+ {
487
+ "step": 25,
488
+ "loss": 0.0,
489
+ "reward": 0.0,
490
+ "zvf": 1.0,
491
+ "gu": 0.0
492
+ },
493
+ {
494
+ "step": 26,
495
+ "loss": 0.0,
496
+ "reward": 0.0,
497
+ "zvf": 1.0,
498
+ "gu": 0.0
499
+ },
500
+ {
501
+ "step": 27,
502
+ "loss": 0.0,
503
+ "reward": 0.0,
504
+ "zvf": 1.0,
505
+ "gu": 0.0
506
+ },
507
+ {
508
+ "step": 28,
509
+ "loss": 0.0,
510
+ "reward": 0.0,
511
+ "zvf": 1.0,
512
+ "gu": 0.0
513
+ },
514
+ {
515
+ "step": 29,
516
+ "loss": 0.0,
517
+ "reward": 0.0,
518
+ "zvf": 1.0,
519
+ "gu": 0.0
520
+ },
521
+ {
522
+ "step": 30,
523
+ "loss": 0.0,
524
+ "reward": 0.0,
525
+ "zvf": 1.0,
526
+ "gu": 0.0
527
+ }
528
+ ],
529
+ "timestamp": "2026-04-19T03:41:34.408077",
530
+ "platform": "tinker",
531
+ "peak_accuracy": 0.0,
532
+ "last10_accuracy": 0.0
533
+ },
534
+ {
535
+ "experiment": "frontier_gsm8k_deepseek-v3.1",
536
+ "model": "deepseek-ai/DeepSeek-V3.1",
537
+ "model_short": "deepseek-v3.1",
538
+ "task": "gsm8k",
539
+ "seed": 42,
540
+ "rank": 16,
541
+ "lr": 1e-05,
542
+ "group": 4,
543
+ "steps": 20,
544
+ "run_id": "56b99b24-03bd-5d00-ae23-831933ef53b2:train:0",
545
+ "checkpoint": "tinker://56b99b24-03bd-5d00-ae23-831933ef53b2:train:0/sampler_weights/final",
546
+ "first5_avg": 0.85,
547
+ "last10_avg": 0.85,
548
+ "peak": 1.0,
549
+ "zero_loss_pct": 30.0,
550
+ "zero_reward_pct": 0.0,
551
+ "reward_trace": [
552
+ 0.875,
553
+ 0.875,
554
+ 1.0,
555
+ 0.75,
556
+ 0.75,
557
+ 0.75,
558
+ 0.625,
559
+ 0.875,
560
+ 0.875,
561
+ 1.0,
562
+ 0.75,
563
+ 1.0,
564
+ 1.0,
565
+ 0.875,
566
+ 1.0,
567
+ 0.5,
568
+ 0.875,
569
+ 0.75,
570
+ 0.75,
571
+ 1.0
572
+ ],
573
+ "step_log": [
574
+ {
575
+ "step": 1,
576
+ "loss": -9.515472412109375,
577
+ "reward": 0.875,
578
+ "zvf": 0.0,
579
+ "gu": 1.0
580
+ },
581
+ {
582
+ "step": 2,
583
+ "loss": -9.95474624633789,
584
+ "reward": 0.875,
585
+ "zvf": 0.0,
586
+ "gu": 1.0
587
+ },
588
+ {
589
+ "step": 3,
590
+ "loss": 0.0,
591
+ "reward": 1.0,
592
+ "zvf": 1.0,
593
+ "gu": 0.0
594
+ },
595
+ {
596
+ "step": 4,
597
+ "loss": -8.243820190429688,
598
+ "reward": 0.75,
599
+ "zvf": 0.0,
600
+ "gu": 1.0
601
+ },
602
+ {
603
+ "step": 5,
604
+ "loss": -4.6991777420043945,
605
+ "reward": 0.75,
606
+ "zvf": 0.0,
607
+ "gu": 1.0
608
+ },
609
+ {
610
+ "step": 6,
611
+ "loss": -11.01620864868164,
612
+ "reward": 0.75,
613
+ "zvf": 0.0,
614
+ "gu": 1.0
615
+ },
616
+ {
617
+ "step": 7,
618
+ "loss": 7.589530944824219,
619
+ "reward": 0.625,
620
+ "zvf": 0.0,
621
+ "gu": 1.0
622
+ },
623
+ {
624
+ "step": 8,
625
+ "loss": -16.8707275390625,
626
+ "reward": 0.875,
627
+ "zvf": 0.0,
628
+ "gu": 1.0
629
+ },
630
+ {
631
+ "step": 9,
632
+ "loss": -13.780792236328125,
633
+ "reward": 0.875,
634
+ "zvf": 0.0,
635
+ "gu": 1.0
636
+ },
637
+ {
638
+ "step": 10,
639
+ "loss": 0.0,
640
+ "reward": 1.0,
641
+ "zvf": 1.0,
642
+ "gu": 0.0
643
+ },
644
+ {
645
+ "step": 11,
646
+ "loss": -19.474306106567383,
647
+ "reward": 0.75,
648
+ "zvf": 0.0,
649
+ "gu": 1.0
650
+ },
651
+ {
652
+ "step": 12,
653
+ "loss": 0.0,
654
+ "reward": 1.0,
655
+ "zvf": 1.0,
656
+ "gu": 0.0
657
+ },
658
+ {
659
+ "step": 13,
660
+ "loss": 0.0,
661
+ "reward": 1.0,
662
+ "zvf": 1.0,
663
+ "gu": 0.0
664
+ },
665
+ {
666
+ "step": 14,
667
+ "loss": -12.405517578125,
668
+ "reward": 0.875,
669
+ "zvf": 0.0,
670
+ "gu": 1.0
671
+ },
672
+ {
673
+ "step": 15,
674
+ "loss": 0.0,
675
+ "reward": 1.0,
676
+ "zvf": 1.0,
677
+ "gu": 0.0
678
+ },
679
+ {
680
+ "step": 16,
681
+ "loss": -14.82606315612793,
682
+ "reward": 0.5,
683
+ "zvf": 0.0,
684
+ "gu": 1.0
685
+ },
686
+ {
687
+ "step": 17,
688
+ "loss": -12.409809112548828,
689
+ "reward": 0.875,
690
+ "zvf": 0.0,
691
+ "gu": 1.0
692
+ },
693
+ {
694
+ "step": 18,
695
+ "loss": -13.506965637207031,
696
+ "reward": 0.75,
697
+ "zvf": 0.0,
698
+ "gu": 1.0
699
+ },
700
+ {
701
+ "step": 19,
702
+ "loss": -14.026712417602539,
703
+ "reward": 0.75,
704
+ "zvf": 0.0,
705
+ "gu": 1.0
706
+ },
707
+ {
708
+ "step": 20,
709
+ "loss": 0.0,
710
+ "reward": 1.0,
711
+ "zvf": 1.0,
712
+ "gu": 0.0
713
+ }
714
+ ],
715
+ "timestamp": "2026-04-18T16:05:46.209173",
716
+ "platform": "tinker",
717
+ "peak_accuracy": 1.0,
718
+ "last10_accuracy": 0.85
719
+ },
720
+ {
721
+ "experiment": "frontier_gsm8k_nemotron-120b",
722
+ "model": "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16",
723
+ "model_short": "nemotron-120b",
724
+ "task": "gsm8k",
725
+ "seed": 42,
726
+ "rank": 16,
727
+ "lr": 1e-05,
728
+ "group": 4,
729
+ "steps": 20,
730
+ "run_id": "657a920a-9e74-55d2-9354-71a6ec2f1f61:train:0",
731
+ "checkpoint": "tinker://657a920a-9e74-55d2-9354-71a6ec2f1f61:train:0/sampler_weights/final",
732
+ "first5_avg": 0.175,
733
+ "last10_avg": 0.1625,
734
+ "peak": 0.875,
735
+ "zero_loss_pct": 70.0,
736
+ "zero_reward_pct": 55.00000000000001,
737
+ "reward_trace": [
738
+ 0.0,
739
+ 0.0,
740
+ 0.875,
741
+ 0.0,
742
+ 0.0,
743
+ 0.0,
744
+ 0.125,
745
+ 0.5,
746
+ 0.375,
747
+ 0.0,
748
+ 0.125,
749
+ 0.25,
750
+ 0.0,
751
+ 0.0,
752
+ 0.25,
753
+ 0.0,
754
+ 0.0,
755
+ 0.0,
756
+ 0.5,
757
+ 0.5
758
+ ],
759
+ "step_log": [
760
+ {
761
+ "step": 1,
762
+ "loss": 0.0,
763
+ "reward": 0.0,
764
+ "zvf": 1.0,
765
+ "gu": 0.0
766
+ },
767
+ {
768
+ "step": 2,
769
+ "loss": 0.0,
770
+ "reward": 0.0,
771
+ "zvf": 1.0,
772
+ "gu": 0.0
773
+ },
774
+ {
775
+ "step": 3,
776
+ "loss": -42.757484436035156,
777
+ "reward": 0.875,
778
+ "zvf": 0.0,
779
+ "gu": 1.0
780
+ },
781
+ {
782
+ "step": 4,
783
+ "loss": 0.0,
784
+ "reward": 0.0,
785
+ "zvf": 1.0,
786
+ "gu": 0.0
787
+ },
788
+ {
789
+ "step": 5,
790
+ "loss": 0.0,
791
+ "reward": 0.0,
792
+ "zvf": 1.0,
793
+ "gu": 0.0
794
+ },
795
+ {
796
+ "step": 6,
797
+ "loss": 0.0,
798
+ "reward": 0.0,
799
+ "zvf": 1.0,
800
+ "gu": 0.0
801
+ },
802
+ {
803
+ "step": 7,
804
+ "loss": 5.0445404052734375,
805
+ "reward": 0.125,
806
+ "zvf": 0.0,
807
+ "gu": 1.0
808
+ },
809
+ {
810
+ "step": 8,
811
+ "loss": 0.0,
812
+ "reward": 0.5,
813
+ "zvf": 0.0,
814
+ "gu": 1.0
815
+ },
816
+ {
817
+ "step": 9,
818
+ "loss": -100.07317352294922,
819
+ "reward": 0.375,
820
+ "zvf": 0.0,
821
+ "gu": 1.0
822
+ },
823
+ {
824
+ "step": 10,
825
+ "loss": 0.0,
826
+ "reward": 0.0,
827
+ "zvf": 1.0,
828
+ "gu": 0.0
829
+ },
830
+ {
831
+ "step": 11,
832
+ "loss": -14.067096710205078,
833
+ "reward": 0.125,
834
+ "zvf": 0.0,
835
+ "gu": 1.0
836
+ },
837
+ {
838
+ "step": 12,
839
+ "loss": -76.11570739746094,
840
+ "reward": 0.25,
841
+ "zvf": 0.0,
842
+ "gu": 1.0
843
+ },
844
+ {
845
+ "step": 13,
846
+ "loss": 0.0,
847
+ "reward": 0.0,
848
+ "zvf": 1.0,
849
+ "gu": 0.0
850
+ },
851
+ {
852
+ "step": 14,
853
+ "loss": 0.0,
854
+ "reward": 0.0,
855
+ "zvf": 1.0,
856
+ "gu": 0.0
857
+ },
858
+ {
859
+ "step": 15,
860
+ "loss": 70.30506896972656,
861
+ "reward": 0.25,
862
+ "zvf": 0.0,
863
+ "gu": 1.0
864
+ },
865
+ {
866
+ "step": 16,
867
+ "loss": 0.0,
868
+ "reward": 0.0,
869
+ "zvf": 1.0,
870
+ "gu": 0.0
871
+ },
872
+ {
873
+ "step": 17,
874
+ "loss": 0.0,
875
+ "reward": 0.0,
876
+ "zvf": 1.0,
877
+ "gu": 0.0
878
+ },
879
+ {
880
+ "step": 18,
881
+ "loss": 0.0,
882
+ "reward": 0.0,
883
+ "zvf": 1.0,
884
+ "gu": 0.0
885
+ },
886
+ {
887
+ "step": 19,
888
+ "loss": 0.0,
889
+ "reward": 0.5,
890
+ "zvf": 0.0,
891
+ "gu": 1.0
892
+ },
893
+ {
894
+ "step": 20,
895
+ "loss": 0.0,
896
+ "reward": 0.5,
897
+ "zvf": 0.0,
898
+ "gu": 1.0
899
+ }
900
+ ],
901
+ "timestamp": "2026-04-19T03:31:55.848278",
902
+ "platform": "tinker",
903
+ "peak_accuracy": 0.875,
904
+ "last10_accuracy": 0.1625
905
+ },
906
+ {
907
+ "experiment": "frontier_gsm8k_qwen3-235b",
908
+ "model": "Qwen/Qwen3-235B-A22B-Instruct-2507",
909
+ "model_short": "qwen3-235b-moe",
910
+ "task": "gsm8k",
911
+ "seed": 42,
912
+ "rank": 16,
913
+ "lr": 1e-05,
914
+ "group": 4,
915
+ "steps": 20,
916
+ "steps_completed": 15,
917
+ "partial": true,
918
+ "run_id": "partial-from-console",
919
+ "checkpoint": "training_interrupted",
920
+ "first5_avg": 1.0,
921
+ "last10_avg": 1.0,
922
+ "peak": 1.0,
923
+ "zero_loss_pct": 0,
924
+ "zero_reward_pct": 0.0,
925
+ "reward_trace": [
926
+ 1.0,
927
+ 1.0,
928
+ 1.0,
929
+ 1.0
930
+ ],
931
+ "platform": "tinker",
932
+ "peak_accuracy": 1.0,
933
+ "last10_accuracy": 1.0
934
+ },
935
+ {
936
+ "experiment": "moe_gsm8k_qwen3-30b-inst",
937
+ "model": "Qwen/Qwen3-30B-A3B-Instruct-2507",
938
+ "model_short": "qwen3-30b-moe-inst",
939
+ "task": "gsm8k",
940
+ "seed": 42,
941
+ "rank": 32,
942
+ "lr": 3e-05,
943
+ "group": 8,
944
+ "steps": 30,
945
+ "steps_completed": 10,
946
+ "partial": true,
947
+ "run_id": "partial-from-console",
948
+ "checkpoint": "training_interrupted",
949
+ "first5_avg": 1.0,
950
+ "last10_avg": 1.0,
951
+ "peak": 1.0,
952
+ "zero_loss_pct": 0,
953
+ "zero_reward_pct": 0.0,
954
+ "reward_trace": [
955
+ 1.0,
956
+ 1.0,
957
+ 1.0
958
+ ],
959
+ "platform": "tinker",
960
+ "peak_accuracy": 1.0,
961
+ "last10_accuracy": 1.0
962
+ },
963
+ {
964
+ "experiment": "moe_gsm8k_qwen3-30b-moe",
965
+ "model": "Qwen/Qwen3-30B-A3B",
966
+ "model_short": "qwen3-30b-moe",
967
+ "task": "gsm8k",
968
+ "seed": 42,
969
+ "rank": 32,
970
+ "lr": 3e-05,
971
+ "group": 8,
972
+ "steps": 30,
973
+ "steps_completed": 20,
974
+ "partial": true,
975
+ "run_id": "partial-from-console",
976
+ "checkpoint": "training_interrupted",
977
+ "first5_avg": 0.3252,
978
+ "last10_avg": 0.3252,
979
+ "peak": 0.5,
980
+ "zero_loss_pct": 0,
981
+ "zero_reward_pct": 0.0,
982
+ "reward_trace": [
983
+ 0.188,
984
+ 0.438,
985
+ 0.062,
986
+ 0.5,
987
+ 0.438
988
+ ],
989
+ "platform": "tinker",
990
+ "peak_accuracy": 0.5,
991
+ "last10_accuracy": 0.3252
992
+ },
993
+ {
994
+ "experiment": "scale_gsm8k_llama-8b-inst",
995
+ "model": "meta-llama/Llama-3.1-8B-Instruct",
996
+ "model_short": "llama-8b-inst",
997
+ "task": "gsm8k",
998
+ "seed": 42,
999
+ "rank": 32,
1000
+ "lr": 3e-05,
1001
+ "group": 8,
1002
+ "steps": 30,
1003
+ "run_id": "488bbb2e-aa35-5431-b4a5-793f68b6bc49:train:0",
1004
+ "checkpoint": "tinker://488bbb2e-aa35-5431-b4a5-793f68b6bc49:train:0/sampler_weights/final",
1005
+ "first5_avg": 0.975,
1006
+ "last10_avg": 0.84375,
1007
+ "peak": 1.0,
1008
+ "zero_loss_pct": 43.333333333333336,
1009
+ "zero_reward_pct": 0.0,
1010
+ "reward_trace": [
1011
+ 1.0,
1012
+ 0.875,
1013
+ 1.0,
1014
+ 1.0,
1015
+ 1.0,
1016
+ 0.625,
1017
+ 1.0,
1018
+ 1.0,
1019
+ 0.875,
1020
+ 0.875,
1021
+ 1.0,
1022
+ 1.0,
1023
+ 0.9375,
1024
+ 0.75,
1025
+ 1.0,
1026
+ 0.375,
1027
+ 0.8125,
1028
+ 0.875,
1029
+ 0.875,
1030
+ 0.75,
1031
+ 0.875,
1032
+ 0.6875,
1033
+ 0.75,
1034
+ 0.4375,
1035
+ 1.0,
1036
+ 1.0,
1037
+ 0.8125,
1038
+ 1.0,
1039
+ 0.875,
1040
+ 1.0
1041
+ ],
1042
+ "step_log": [
1043
+ {
1044
+ "step": 1,
1045
+ "loss": 0.0,
1046
+ "reward": 1.0,
1047
+ "zvf": 1.0,
1048
+ "gu": 0.0
1049
+ },
1050
+ {
1051
+ "step": 2,
1052
+ "loss": 6.620754241943359,
1053
+ "reward": 0.875,
1054
+ "zvf": 0.0,
1055
+ "gu": 1.0
1056
+ },
1057
+ {
1058
+ "step": 3,
1059
+ "loss": 0.0,
1060
+ "reward": 1.0,
1061
+ "zvf": 1.0,
1062
+ "gu": 0.0
1063
+ },
1064
+ {
1065
+ "step": 4,
1066
+ "loss": 0.0,
1067
+ "reward": 1.0,
1068
+ "zvf": 1.0,
1069
+ "gu": 0.0
1070
+ },
1071
+ {
1072
+ "step": 5,
1073
+ "loss": 0.0,
1074
+ "reward": 1.0,
1075
+ "zvf": 1.0,
1076
+ "gu": 0.0
1077
+ },
1078
+ {
1079
+ "step": 6,
1080
+ "loss": 2.9265289306640625,
1081
+ "reward": 0.625,
1082
+ "zvf": 0.0,
1083
+ "gu": 1.0
1084
+ },
1085
+ {
1086
+ "step": 7,
1087
+ "loss": 0.0,
1088
+ "reward": 1.0,
1089
+ "zvf": 1.0,
1090
+ "gu": 0.0
1091
+ },
1092
+ {
1093
+ "step": 8,
1094
+ "loss": 0.0,
1095
+ "reward": 1.0,
1096
+ "zvf": 1.0,
1097
+ "gu": 0.0
1098
+ },
1099
+ {
1100
+ "step": 9,
1101
+ "loss": -0.9360866546630859,
1102
+ "reward": 0.875,
1103
+ "zvf": 0.0,
1104
+ "gu": 1.0
1105
+ },
1106
+ {
1107
+ "step": 10,
1108
+ "loss": -2.864093780517578,
1109
+ "reward": 0.875,
1110
+ "zvf": 0.0,
1111
+ "gu": 1.0
1112
+ },
1113
+ {
1114
+ "step": 11,
1115
+ "loss": 0.0,
1116
+ "reward": 1.0,
1117
+ "zvf": 1.0,
1118
+ "gu": 0.0
1119
+ },
1120
+ {
1121
+ "step": 12,
1122
+ "loss": 0.0,
1123
+ "reward": 1.0,
1124
+ "zvf": 1.0,
1125
+ "gu": 0.0
1126
+ },
1127
+ {
1128
+ "step": 13,
1129
+ "loss": -22.385391235351562,
1130
+ "reward": 0.9375,
1131
+ "zvf": 0.0,
1132
+ "gu": 1.0
1133
+ },
1134
+ {
1135
+ "step": 14,
1136
+ "loss": -5.675556182861328,
1137
+ "reward": 0.75,
1138
+ "zvf": 0.0,
1139
+ "gu": 1.0
1140
+ },
1141
+ {
1142
+ "step": 15,
1143
+ "loss": 0.0,
1144
+ "reward": 1.0,
1145
+ "zvf": 1.0,
1146
+ "gu": 0.0
1147
+ },
1148
+ {
1149
+ "step": 16,
1150
+ "loss": 7.295082092285156,
1151
+ "reward": 0.375,
1152
+ "zvf": 0.0,
1153
+ "gu": 1.0
1154
+ },
1155
+ {
1156
+ "step": 17,
1157
+ "loss": 2.713848114013672,
1158
+ "reward": 0.8125,
1159
+ "zvf": 0.0,
1160
+ "gu": 1.0
1161
+ },
1162
+ {
1163
+ "step": 18,
1164
+ "loss": -18.252927780151367,
1165
+ "reward": 0.875,
1166
+ "zvf": 0.0,
1167
+ "gu": 1.0
1168
+ },
1169
+ {
1170
+ "step": 19,
1171
+ "loss": -5.822566986083984,
1172
+ "reward": 0.875,
1173
+ "zvf": 0.0,
1174
+ "gu": 1.0
1175
+ },
1176
+ {
1177
+ "step": 20,
1178
+ "loss": 8.464057922363281,
1179
+ "reward": 0.75,
1180
+ "zvf": 0.0,
1181
+ "gu": 1.0
1182
+ },
1183
+ {
1184
+ "step": 21,
1185
+ "loss": 13.122426986694336,
1186
+ "reward": 0.875,
1187
+ "zvf": 0.0,
1188
+ "gu": 1.0
1189
+ },
1190
+ {
1191
+ "step": 22,
1192
+ "loss": -19.496044158935547,
1193
+ "reward": 0.6875,
1194
+ "zvf": 0.0,
1195
+ "gu": 1.0
1196
+ },
1197
+ {
1198
+ "step": 23,
1199
+ "loss": 0.4438667297363281,
1200
+ "reward": 0.75,
1201
+ "zvf": 0.0,
1202
+ "gu": 1.0
1203
+ },
1204
+ {
1205
+ "step": 24,
1206
+ "loss": 4.036277770996094,
1207
+ "reward": 0.4375,
1208
+ "zvf": 0.0,
1209
+ "gu": 1.0
1210
+ },
1211
+ {
1212
+ "step": 25,
1213
+ "loss": 0.0,
1214
+ "reward": 1.0,
1215
+ "zvf": 1.0,
1216
+ "gu": 0.0
1217
+ },
1218
+ {
1219
+ "step": 26,
1220
+ "loss": 0.0,
1221
+ "reward": 1.0,
1222
+ "zvf": 1.0,
1223
+ "gu": 0.0
1224
+ },
1225
+ {
1226
+ "step": 27,
1227
+ "loss": 3.3459701538085938,
1228
+ "reward": 0.8125,
1229
+ "zvf": 0.0,
1230
+ "gu": 1.0
1231
+ },
1232
+ {
1233
+ "step": 28,
1234
+ "loss": 0.0,
1235
+ "reward": 1.0,
1236
+ "zvf": 1.0,
1237
+ "gu": 0.0
1238
+ },
1239
+ {
1240
+ "step": 29,
1241
+ "loss": 12.346473693847656,
1242
+ "reward": 0.875,
1243
+ "zvf": 0.0,
1244
+ "gu": 1.0
1245
+ },
1246
+ {
1247
+ "step": 30,
1248
+ "loss": 0.0,
1249
+ "reward": 1.0,
1250
+ "zvf": 1.0,
1251
+ "gu": 0.0
1252
+ }
1253
+ ],
1254
+ "timestamp": "2026-04-19T04:11:44.481966",
1255
+ "platform": "tinker",
1256
+ "peak_accuracy": 1.0,
1257
+ "last10_accuracy": 0.84375
1258
+ },
1259
+ {
1260
+ "experiment": "scale_gsm8k_qwen3-32b",
1261
+ "model": "Qwen/Qwen3-32B",
1262
+ "model_short": "qwen3-32b",
1263
+ "task": "gsm8k",
1264
+ "seed": 42,
1265
+ "rank": 32,
1266
+ "lr": 3e-05,
1267
+ "group": 8,
1268
+ "steps": 30,
1269
+ "steps_completed": 10,
1270
+ "partial": true,
1271
+ "run_id": "partial-from-console",
1272
+ "checkpoint": "training_interrupted",
1273
+ "first5_avg": 0.24966666666666668,
1274
+ "last10_avg": 0.24966666666666668,
1275
+ "peak": 0.312,
1276
+ "zero_loss_pct": 0,
1277
+ "zero_reward_pct": 0.0,
1278
+ "reward_trace": [
1279
+ 0.312,
1280
+ 0.312,
1281
+ 0.125
1282
+ ],
1283
+ "platform": "tinker",
1284
+ "peak_accuracy": 0.312,
1285
+ "last10_accuracy": 0.24966666666666668
1286
+ },
1287
+ {
1288
+ "experiment": "scale_gsm8k_qwen3-8b",
1289
+ "model": "Qwen/Qwen3-8B",
1290
+ "model_short": "qwen3-8b",
1291
+ "task": "gsm8k",
1292
+ "seed": 42,
1293
+ "rank": 32,
1294
+ "lr": 3e-05,
1295
+ "group": 8,
1296
+ "steps": 30,
1297
+ "run_id": "3154dee5-4fd7-59b6-ba3b-721eef675bfc:train:0",
1298
+ "checkpoint": "tinker://3154dee5-4fd7-59b6-ba3b-721eef675bfc:train:0/sampler_weights/final",
1299
+ "first5_avg": 0.3125,
1300
+ "last10_avg": 0.34375,
1301
+ "peak": 0.625,
1302
+ "zero_loss_pct": 6.666666666666667,
1303
+ "zero_reward_pct": 6.666666666666667,
1304
+ "reward_trace": [
1305
+ 0.375,
1306
+ 0.0625,
1307
+ 0.5625,
1308
+ 0.1875,
1309
+ 0.375,
1310
+ 0.3125,
1311
+ 0.125,
1312
+ 0.125,
1313
+ 0.1875,
1314
+ 0.125,
1315
+ 0.25,
1316
+ 0.4375,
1317
+ 0.625,
1318
+ 0.0,
1319
+ 0.4375,
1320
+ 0.0625,
1321
+ 0.1875,
1322
+ 0.0,
1323
+ 0.375,
1324
+ 0.3125,
1325
+ 0.3125,
1326
+ 0.1875,
1327
+ 0.375,
1328
+ 0.375,
1329
+ 0.5,
1330
+ 0.25,
1331
+ 0.4375,
1332
+ 0.1875,
1333
+ 0.625,
1334
+ 0.1875
1335
+ ],
1336
+ "step_log": [
1337
+ {
1338
+ "step": 1,
1339
+ "loss": -3.936065673828125,
1340
+ "reward": 0.375,
1341
+ "zvf": 0.0,
1342
+ "gu": 1.0
1343
+ },
1344
+ {
1345
+ "step": 2,
1346
+ "loss": -2.969602584838867,
1347
+ "reward": 0.0625,
1348
+ "zvf": 0.0,
1349
+ "gu": 1.0
1350
+ },
1351
+ {
1352
+ "step": 3,
1353
+ "loss": -4.0252532958984375,
1354
+ "reward": 0.5625,
1355
+ "zvf": 0.0,
1356
+ "gu": 1.0
1357
+ },
1358
+ {
1359
+ "step": 4,
1360
+ "loss": -3.1384735107421875,
1361
+ "reward": 0.1875,
1362
+ "zvf": 0.0,
1363
+ "gu": 1.0
1364
+ },
1365
+ {
1366
+ "step": 5,
1367
+ "loss": -5.50750732421875,
1368
+ "reward": 0.375,
1369
+ "zvf": 0.0,
1370
+ "gu": 1.0
1371
+ },
1372
+ {
1373
+ "step": 6,
1374
+ "loss": -2.173931121826172,
1375
+ "reward": 0.3125,
1376
+ "zvf": 0.0,
1377
+ "gu": 1.0
1378
+ },
1379
+ {
1380
+ "step": 7,
1381
+ "loss": -2.5752410888671875,
1382
+ "reward": 0.125,
1383
+ "zvf": 0.0,
1384
+ "gu": 1.0
1385
+ },
1386
+ {
1387
+ "step": 8,
1388
+ "loss": -2.2352685928344727,
1389
+ "reward": 0.125,
1390
+ "zvf": 0.0,
1391
+ "gu": 1.0
1392
+ },
1393
+ {
1394
+ "step": 9,
1395
+ "loss": -4.3625383377075195,
1396
+ "reward": 0.1875,
1397
+ "zvf": 0.0,
1398
+ "gu": 1.0
1399
+ },
1400
+ {
1401
+ "step": 10,
1402
+ "loss": -1.4611568450927734,
1403
+ "reward": 0.125,
1404
+ "zvf": 0.0,
1405
+ "gu": 1.0
1406
+ },
1407
+ {
1408
+ "step": 11,
1409
+ "loss": -8.721845626831055,
1410
+ "reward": 0.25,
1411
+ "zvf": 0.0,
1412
+ "gu": 1.0
1413
+ },
1414
+ {
1415
+ "step": 12,
1416
+ "loss": 1.46728515625,
1417
+ "reward": 0.4375,
1418
+ "zvf": 0.0,
1419
+ "gu": 1.0
1420
+ },
1421
+ {
1422
+ "step": 13,
1423
+ "loss": -6.392450332641602,
1424
+ "reward": 0.625,
1425
+ "zvf": 0.0,
1426
+ "gu": 1.0
1427
+ },
1428
+ {
1429
+ "step": 14,
1430
+ "loss": 0.0,
1431
+ "reward": 0.0,
1432
+ "zvf": 1.0,
1433
+ "gu": 0.0
1434
+ },
1435
+ {
1436
+ "step": 15,
1437
+ "loss": -2.3406524658203125,
1438
+ "reward": 0.4375,
1439
+ "zvf": 0.0,
1440
+ "gu": 1.0
1441
+ },
1442
+ {
1443
+ "step": 16,
1444
+ "loss": 1.5611343383789062,
1445
+ "reward": 0.0625,
1446
+ "zvf": 0.0,
1447
+ "gu": 1.0
1448
+ },
1449
+ {
1450
+ "step": 17,
1451
+ "loss": -2.0525875091552734,
1452
+ "reward": 0.1875,
1453
+ "zvf": 0.0,
1454
+ "gu": 1.0
1455
+ },
1456
+ {
1457
+ "step": 18,
1458
+ "loss": 0.0,
1459
+ "reward": 0.0,
1460
+ "zvf": 1.0,
1461
+ "gu": 0.0
1462
+ },
1463
+ {
1464
+ "step": 19,
1465
+ "loss": -7.680427551269531,
1466
+ "reward": 0.375,
1467
+ "zvf": 0.0,
1468
+ "gu": 1.0
1469
+ },
1470
+ {
1471
+ "step": 20,
1472
+ "loss": -3.0284852981567383,
1473
+ "reward": 0.3125,
1474
+ "zvf": 0.0,
1475
+ "gu": 1.0
1476
+ },
1477
+ {
1478
+ "step": 21,
1479
+ "loss": -1.5452747344970703,
1480
+ "reward": 0.3125,
1481
+ "zvf": 0.0,
1482
+ "gu": 1.0
1483
+ },
1484
+ {
1485
+ "step": 22,
1486
+ "loss": -2.5575408935546875,
1487
+ "reward": 0.1875,
1488
+ "zvf": 0.0,
1489
+ "gu": 1.0
1490
+ },
1491
+ {
1492
+ "step": 23,
1493
+ "loss": -4.197565078735352,
1494
+ "reward": 0.375,
1495
+ "zvf": 0.0,
1496
+ "gu": 1.0
1497
+ },
1498
+ {
1499
+ "step": 24,
1500
+ "loss": -3.269200325012207,
1501
+ "reward": 0.375,
1502
+ "zvf": 0.0,
1503
+ "gu": 1.0
1504
+ },
1505
+ {
1506
+ "step": 25,
1507
+ "loss": -6.419197082519531,
1508
+ "reward": 0.5,
1509
+ "zvf": 0.0,
1510
+ "gu": 1.0
1511
+ },
1512
+ {
1513
+ "step": 26,
1514
+ "loss": -1.5014839172363281,
1515
+ "reward": 0.25,
1516
+ "zvf": 0.0,
1517
+ "gu": 1.0
1518
+ },
1519
+ {
1520
+ "step": 27,
1521
+ "loss": -4.115062713623047,
1522
+ "reward": 0.4375,
1523
+ "zvf": 0.0,
1524
+ "gu": 1.0
1525
+ },
1526
+ {
1527
+ "step": 28,
1528
+ "loss": -2.7128124237060547,
1529
+ "reward": 0.1875,
1530
+ "zvf": 0.0,
1531
+ "gu": 1.0
1532
+ },
1533
+ {
1534
+ "step": 29,
1535
+ "loss": -2.6960220336914062,
1536
+ "reward": 0.625,
1537
+ "zvf": 0.0,
1538
+ "gu": 1.0
1539
+ },
1540
+ {
1541
+ "step": 30,
1542
+ "loss": -0.14708518981933594,
1543
+ "reward": 0.1875,
1544
+ "zvf": 0.0,
1545
+ "gu": 1.0
1546
+ }
1547
+ ],
1548
+ "timestamp": "2026-04-18T16:31:17.561535",
1549
+ "platform": "tinker",
1550
+ "peak_accuracy": 0.625,
1551
+ "last10_accuracy": 0.34375
1552
+ },
1553
+ {
1554
+ "experiment": "scale_gsm8k_qwen3.5-27b",
1555
+ "model": "Qwen/Qwen3.5-27B",
1556
+ "model_short": "qwen3.5-27b",
1557
+ "task": "gsm8k",
1558
+ "seed": 42,
1559
+ "rank": 32,
1560
+ "lr": 3e-05,
1561
+ "group": 8,
1562
+ "steps": 30,
1563
+ "steps_completed": 10,
1564
+ "partial": true,
1565
+ "run_id": "partial-from-console",
1566
+ "checkpoint": "training_interrupted",
1567
+ "first5_avg": 0.43733333333333335,
1568
+ "last10_avg": 0.43733333333333335,
1569
+ "peak": 0.75,
1570
+ "zero_loss_pct": 0,
1571
+ "zero_reward_pct": 33.33333333333333,
1572
+ "reward_trace": [
1573
+ 0.562,
1574
+ 0.75,
1575
+ 0.0
1576
+ ],
1577
+ "platform": "tinker",
1578
+ "peak_accuracy": 0.75,
1579
+ "last10_accuracy": 0.43733333333333335
1580
+ },
1581
+ {
1582
+ "experiment": "scale_gsm8k_qwen3.5-4b",
1583
+ "model": "Qwen/Qwen3.5-4B",
1584
+ "model_short": "qwen3.5-4b",
1585
+ "task": "gsm8k",
1586
+ "seed": 42,
1587
+ "rank": 32,
1588
+ "lr": 3e-05,
1589
+ "group": 8,
1590
+ "steps": 30,
1591
+ "run_id": "200e4983-3e9d-5176-87d0-c55dd3c73539:train:0",
1592
+ "checkpoint": "tinker://200e4983-3e9d-5176-87d0-c55dd3c73539:train:0/sampler_weights/final",
1593
+ "first5_avg": 0.9625,
1594
+ "last10_avg": 0.85,
1595
+ "peak": 1.0,
1596
+ "zero_loss_pct": 56.666666666666664,
1597
+ "zero_reward_pct": 0.0,
1598
+ "reward_trace": [
1599
+ 1.0,
1600
+ 0.8125,
1601
+ 1.0,
1602
+ 1.0,
1603
+ 1.0,
1604
+ 0.5,
1605
+ 0.5,
1606
+ 0.875,
1607
+ 0.625,
1608
+ 0.8125,
1609
+ 1.0,
1610
+ 1.0,
1611
+ 1.0,
1612
+ 0.625,
1613
+ 0.875,
1614
+ 0.25,
1615
+ 1.0,
1616
+ 0.875,
1617
+ 0.6875,
1618
+ 0.5625,
1619
+ 0.9375,
1620
+ 0.5,
1621
+ 0.9375,
1622
+ 0.5,
1623
+ 1.0,
1624
+ 1.0,
1625
+ 0.625,
1626
+ 1.0,
1627
+ 1.0,
1628
+ 1.0
1629
+ ],
1630
+ "step_log": [
1631
+ {
1632
+ "step": 1,
1633
+ "loss": 0.0,
1634
+ "reward": 1.0,
1635
+ "zvf": 1.0,
1636
+ "gu": 0.0
1637
+ },
1638
+ {
1639
+ "step": 2,
1640
+ "loss": -2.847966194152832,
1641
+ "reward": 0.8125,
1642
+ "zvf": 0.0,
1643
+ "gu": 1.0
1644
+ },
1645
+ {
1646
+ "step": 3,
1647
+ "loss": 0.0,
1648
+ "reward": 1.0,
1649
+ "zvf": 1.0,
1650
+ "gu": 0.0
1651
+ },
1652
+ {
1653
+ "step": 4,
1654
+ "loss": 0.0,
1655
+ "reward": 1.0,
1656
+ "zvf": 1.0,
1657
+ "gu": 0.0
1658
+ },
1659
+ {
1660
+ "step": 5,
1661
+ "loss": 0.0,
1662
+ "reward": 1.0,
1663
+ "zvf": 1.0,
1664
+ "gu": 0.0
1665
+ },
1666
+ {
1667
+ "step": 6,
1668
+ "loss": 0.0,
1669
+ "reward": 0.5,
1670
+ "zvf": 0.0,
1671
+ "gu": 1.0
1672
+ },
1673
+ {
1674
+ "step": 7,
1675
+ "loss": 0.0,
1676
+ "reward": 0.5,
1677
+ "zvf": 0.0,
1678
+ "gu": 1.0
1679
+ },
1680
+ {
1681
+ "step": 8,
1682
+ "loss": -1.0574636459350586,
1683
+ "reward": 0.875,
1684
+ "zvf": 0.0,
1685
+ "gu": 1.0
1686
+ },
1687
+ {
1688
+ "step": 9,
1689
+ "loss": -1.7392759323120117,
1690
+ "reward": 0.625,
1691
+ "zvf": 0.0,
1692
+ "gu": 1.0
1693
+ },
1694
+ {
1695
+ "step": 10,
1696
+ "loss": -23.38077163696289,
1697
+ "reward": 0.8125,
1698
+ "zvf": 0.0,
1699
+ "gu": 1.0
1700
+ },
1701
+ {
1702
+ "step": 11,
1703
+ "loss": 0.0,
1704
+ "reward": 1.0,
1705
+ "zvf": 1.0,
1706
+ "gu": 0.0
1707
+ },
1708
+ {
1709
+ "step": 12,
1710
+ "loss": 0.0,
1711
+ "reward": 1.0,
1712
+ "zvf": 1.0,
1713
+ "gu": 0.0
1714
+ },
1715
+ {
1716
+ "step": 13,
1717
+ "loss": 0.0,
1718
+ "reward": 1.0,
1719
+ "zvf": 1.0,
1720
+ "gu": 0.0
1721
+ },
1722
+ {
1723
+ "step": 14,
1724
+ "loss": -5.09890079498291,
1725
+ "reward": 0.625,
1726
+ "zvf": 0.0,
1727
+ "gu": 1.0
1728
+ },
1729
+ {
1730
+ "step": 15,
1731
+ "loss": -17.38865089416504,
1732
+ "reward": 0.875,
1733
+ "zvf": 0.0,
1734
+ "gu": 1.0
1735
+ },
1736
+ {
1737
+ "step": 16,
1738
+ "loss": -10.52424430847168,
1739
+ "reward": 0.25,
1740
+ "zvf": 0.0,
1741
+ "gu": 1.0
1742
+ },
1743
+ {
1744
+ "step": 17,
1745
+ "loss": 0.0,
1746
+ "reward": 1.0,
1747
+ "zvf": 1.0,
1748
+ "gu": 0.0
1749
+ },
1750
+ {
1751
+ "step": 18,
1752
+ "loss": -12.190226554870605,
1753
+ "reward": 0.875,
1754
+ "zvf": 0.0,
1755
+ "gu": 1.0
1756
+ },
1757
+ {
1758
+ "step": 19,
1759
+ "loss": -6.337299346923828,
1760
+ "reward": 0.6875,
1761
+ "zvf": 0.0,
1762
+ "gu": 1.0
1763
+ },
1764
+ {
1765
+ "step": 20,
1766
+ "loss": -4.9612884521484375,
1767
+ "reward": 0.5625,
1768
+ "zvf": 0.0,
1769
+ "gu": 1.0
1770
+ },
1771
+ {
1772
+ "step": 21,
1773
+ "loss": -10.336174011230469,
1774
+ "reward": 0.9375,
1775
+ "zvf": 0.0,
1776
+ "gu": 1.0
1777
+ },
1778
+ {
1779
+ "step": 22,
1780
+ "loss": 0.0,
1781
+ "reward": 0.5,
1782
+ "zvf": 0.0,
1783
+ "gu": 1.0
1784
+ },
1785
+ {
1786
+ "step": 23,
1787
+ "loss": -5.058204650878906,
1788
+ "reward": 0.9375,
1789
+ "zvf": 0.0,
1790
+ "gu": 1.0
1791
+ },
1792
+ {
1793
+ "step": 24,
1794
+ "loss": 0.0,
1795
+ "reward": 0.5,
1796
+ "zvf": 0.0,
1797
+ "gu": 1.0
1798
+ },
1799
+ {
1800
+ "step": 25,
1801
+ "loss": 0.0,
1802
+ "reward": 1.0,
1803
+ "zvf": 1.0,
1804
+ "gu": 0.0
1805
+ },
1806
+ {
1807
+ "step": 26,
1808
+ "loss": 0.0,
1809
+ "reward": 1.0,
1810
+ "zvf": 1.0,
1811
+ "gu": 0.0
1812
+ },
1813
+ {
1814
+ "step": 27,
1815
+ "loss": -7.397833824157715,
1816
+ "reward": 0.625,
1817
+ "zvf": 0.0,
1818
+ "gu": 1.0
1819
+ },
1820
+ {
1821
+ "step": 28,
1822
+ "loss": 0.0,
1823
+ "reward": 1.0,
1824
+ "zvf": 1.0,
1825
+ "gu": 0.0
1826
+ },
1827
+ {
1828
+ "step": 29,
1829
+ "loss": 0.0,
1830
+ "reward": 1.0,
1831
+ "zvf": 1.0,
1832
+ "gu": 0.0
1833
+ },
1834
+ {
1835
+ "step": 30,
1836
+ "loss": 0.0,
1837
+ "reward": 1.0,
1838
+ "zvf": 1.0,
1839
+ "gu": 0.0
1840
+ }
1841
+ ],
1842
+ "timestamp": "2026-04-19T03:34:22.495977",
1843
+ "platform": "tinker",
1844
+ "peak_accuracy": 1.0,
1845
+ "last10_accuracy": 0.85
1846
+ },
1847
+ {
1848
+ "experiment": "kl_track_Qwen3-8B_s42",
1849
+ "model": "Qwen/Qwen3-8B",
1850
+ "model_short": "qwen3-8b",
1851
+ "task": "kl_tracking",
1852
+ "seed": 42,
1853
+ "steps": 30,
1854
+ "final_kl": 60.75,
1855
+ "final_entropy": 0.742,
1856
+ "platform": "modal_h100"
1857
+ },
1858
+ {
1859
+ "experiment": "ppo_gsm8k_qwen3-8b",
1860
+ "model": "Qwen/Qwen3-8B",
1861
+ "model_short": "qwen3-8b",
1862
+ "task": "gsm8k",
1863
+ "seed": 42,
1864
+ "steps": 30,
1865
+ "peak": 0.75,
1866
+ "last10_avg": 0.225,
1867
+ "platform": "modal_h100",
1868
+ "algorithm": "PPO"
1869
+ },
1870
+ {
1871
+ "experiment": "ppo_gsm8k_llama-8b",
1872
+ "model": "meta-llama/Llama-3.1-8B-Instruct",
1873
+ "model_short": "llama-8b-inst",
1874
+ "task": "gsm8k",
1875
+ "seed": 42,
1876
+ "steps": 30,
1877
+ "peak": 1.0,
1878
+ "last10_avg": 0.975,
1879
+ "platform": "modal_h100",
1880
+ "algorithm": "PPO"
1881
+ }
1882
+ ]
arithmetic_checkpoints.jsonl ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ {"name": "000020", "batch": 20, "state_path": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/weights/000020", "sampler_path": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/sampler_weights/000020"}
2
+ {"name": "000040", "batch": 40, "state_path": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/weights/000040", "sampler_path": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/sampler_weights/000040"}
3
+ {"name": "000060", "batch": 60, "state_path": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/weights/000060", "sampler_path": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/sampler_weights/000060"}
4
+ {"name": "000080", "batch": 80, "state_path": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/weights/000080", "sampler_path": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/sampler_weights/000080"}
5
+ {"name": "000100", "batch": 100, "state_path": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/weights/000100", "sampler_path": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/sampler_weights/000100"}
6
+ {"name": "final", "batch": 100, "state_path": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/weights/final", "sampler_path": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/sampler_weights/final"}
checkpoints.jsonl ADDED
@@ -0,0 +1,66 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"tinker_uri": "tinker://c465d654-a00f-5821-9c11-0e40f0eecf9f:train:0/sampler_weights/final", "training_run_id": "c465d654-a00f-5821-9c11-0e40f0eecf9f", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/results/heldout_gsm8k.json", "experiments/tinker-runs/results/arch_gsm8k_gpt-oss-20b.json", "experiments/tinker-runs/results/tinker_parallel_20260419_064834.json"], "experiment": "arch_gsm8k_gpt-oss-20b", "model": "openai/gpt-oss-20b", "model_short": "gpt-oss-20b", "task": "gsm8k", "seed": 42, "rank": 32, "lr": 3e-05, "group": 8, "steps": 30}
2
+ {"tinker_uri": "tinker://51a8ef9e-15ef-5f8f-bda1-78ee51387a12:train:0/sampler_weights/final", "training_run_id": "51a8ef9e-15ef-5f8f-bda1-78ee51387a12", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/tinker-runs/results/arch_gsm8k_kimi-k2.json", "experiments/tinker-runs/results/tinker_parallel_20260419_064834.json"], "experiment": "arch_gsm8k_kimi-k2", "model": "moonshotai/Kimi-K2-Thinking", "model_short": "kimi-k2", "task": "gsm8k", "seed": 42, "rank": 16, "lr": 1e-05, "group": 4, "steps": 20}
3
+ {"tinker_uri": "tinker://ca2e3a24-7401-5770-af34-a0d27177aeaa:train:0/sampler_weights/final", "training_run_id": "ca2e3a24-7401-5770-af34-a0d27177aeaa", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/all_results_consolidated.json", "experiments/master_results.json", "experiments/tinker-runs/results/cross_tool_llama-8b-inst.json"], "experiment": "cross_tool_llama-8b-inst", "model": "meta-llama/Llama-3.1-8B-Instruct", "model_short": "llama-8b-inst", "task": "tool_use", "seed": 42, "rank": 32, "lr": 3e-05, "group": 8, "steps": 30, "platform": "tinker", "last10_avg": 0.0, "peak": 0.0, "peak_accuracy": 0.0, "last10_accuracy": 0.0, "timestamp": "2026-04-18T16:01:52.013977", "source_json": "experiments/master_results.json"}
4
+ {"tinker_uri": "tinker://73ff186a-d8e3-50c3-afb8-2b863cd09579:train:0/sampler_weights/final", "training_run_id": "73ff186a-d8e3-50c3-afb8-2b863cd09579", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/all_results_consolidated.json", "experiments/master_results.json", "experiments/tinker-runs/results/cross_tool_qwen3-32b.json"], "experiment": "cross_tool_qwen3-32b", "model": "Qwen/Qwen3-32B", "model_short": "qwen3-32b", "task": "tool_use", "seed": 42, "rank": 32, "lr": 3e-05, "group": 8, "steps": 30, "platform": "tinker", "last10_avg": 0.0, "peak": 0.0, "peak_accuracy": 0.0, "last10_accuracy": 0.0, "timestamp": "2026-04-19T03:41:34.408077", "source_json": "experiments/master_results.json"}
5
+ {"tinker_uri": "tinker://56b99b24-03bd-5d00-ae23-831933ef53b2:train:0/sampler_weights/final", "training_run_id": "56b99b24-03bd-5d00-ae23-831933ef53b2", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/all_results_consolidated.json", "experiments/master_results.json", "experiments/results/heldout_gsm8k.json", "experiments/tinker-runs/results/frontier_gsm8k_deepseek-v3.1.json"], "experiment": "frontier_gsm8k_deepseek-v3.1", "model": "deepseek-ai/DeepSeek-V3.1", "model_short": "deepseek-v3.1", "task": "gsm8k", "seed": 42, "rank": 16, "lr": 1e-05, "group": 4, "steps": 20, "platform": "tinker", "last10_avg": 0.85, "peak": 1.0, "peak_accuracy": 1.0, "last10_accuracy": 0.85, "timestamp": "2026-04-18T16:05:46.209173", "source_json": "experiments/master_results.json"}
6
+ {"tinker_uri": "tinker://657a920a-9e74-55d2-9354-71a6ec2f1f61:train:0/sampler_weights/final", "training_run_id": "657a920a-9e74-55d2-9354-71a6ec2f1f61", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/all_results_consolidated.json", "experiments/master_results.json", "experiments/tinker-runs/results/frontier_gsm8k_nemotron-120b.json"], "experiment": "frontier_gsm8k_nemotron-120b", "model": "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16", "model_short": "nemotron-120b", "task": "gsm8k", "seed": 42, "rank": 16, "lr": 1e-05, "group": 4, "steps": 20, "platform": "tinker", "last10_avg": 0.1625, "peak": 0.875, "peak_accuracy": 0.875, "last10_accuracy": 0.1625, "timestamp": "2026-04-19T03:31:55.848278", "source_json": "experiments/master_results.json"}
7
+ {"tinker_uri": "tinker://488bbb2e-aa35-5431-b4a5-793f68b6bc49:train:0/sampler_weights/final", "training_run_id": "488bbb2e-aa35-5431-b4a5-793f68b6bc49", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/all_results_consolidated.json", "experiments/master_results.json", "experiments/results/heldout_gsm8k.json", "experiments/tinker-runs/results/scale_gsm8k_llama-8b-inst.json"], "experiment": "scale_gsm8k_llama-8b-inst", "model": "meta-llama/Llama-3.1-8B-Instruct", "model_short": "llama-8b-inst", "task": "gsm8k", "seed": 42, "rank": 32, "lr": 3e-05, "group": 8, "steps": 30, "platform": "tinker", "last10_avg": 0.84375, "peak": 1.0, "peak_accuracy": 1.0, "last10_accuracy": 0.84375, "timestamp": "2026-04-19T04:11:44.481966", "source_json": "experiments/master_results.json"}
8
+ {"tinker_uri": "tinker://3154dee5-4fd7-59b6-ba3b-721eef675bfc:train:0/sampler_weights/final", "training_run_id": "3154dee5-4fd7-59b6-ba3b-721eef675bfc", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/all_results_consolidated.json", "experiments/master_results.json", "experiments/tinker-runs/results/scale_gsm8k_qwen3-8b.json"], "experiment": "scale_gsm8k_qwen3-8b", "model": "Qwen/Qwen3-8B", "model_short": "qwen3-8b", "task": "gsm8k", "seed": 42, "rank": 32, "lr": 3e-05, "group": 8, "steps": 30, "platform": "tinker", "last10_avg": 0.34375, "peak": 0.625, "peak_accuracy": 0.625, "last10_accuracy": 0.34375, "timestamp": "2026-04-18T16:31:17.561535", "source_json": "experiments/master_results.json"}
9
+ {"tinker_uri": "tinker://200e4983-3e9d-5176-87d0-c55dd3c73539:train:0/sampler_weights/final", "training_run_id": "200e4983-3e9d-5176-87d0-c55dd3c73539", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/all_results_consolidated.json", "experiments/master_results.json", "experiments/results/heldout_gsm8k.json", "experiments/tinker-runs/results/scale_gsm8k_qwen3.5-4b.json"], "experiment": "scale_gsm8k_qwen3.5-4b", "model": "Qwen/Qwen3.5-4B", "model_short": "qwen3.5-4b", "task": "gsm8k", "seed": 42, "rank": 32, "lr": 3e-05, "group": 8, "steps": 30, "platform": "tinker", "last10_avg": 0.85, "peak": 1.0, "peak_accuracy": 1.0, "last10_accuracy": 0.85, "timestamp": "2026-04-19T03:34:22.495977", "source_json": "experiments/master_results.json"}
10
+ {"tinker_uri": "tinker://0416d5d1-161b-5264-8b69-4612314d3b41:train:0/sampler_weights/final", "training_run_id": "0416d5d1-161b-5264-8b69-4612314d3b41", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/tinker-runs/results/campaign_v2_20260419_111623.json"], "model": "deepseek-ai/DeepSeek-V3.1", "seed": 456, "lr": 1e-05, "steps": 30}
11
+ {"tinker_uri": "tinker://16d810a0-9818-5f4c-820c-23fc91270769:train:0/sampler_weights/final", "training_run_id": "16d810a0-9818-5f4c-820c-23fc91270769", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/tinker-runs/results/campaign_v2_20260419_111623.json"], "model": "Qwen/Qwen3-8B", "seed": 42, "lr": 1e-05, "steps": 30}
12
+ {"tinker_uri": "tinker://200252fc-0649-5b39-8030-ba0fa5d07e84:train:0/sampler_weights/final", "training_run_id": "200252fc-0649-5b39-8030-ba0fa5d07e84", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/tinker-runs/results/campaign_v2_20260419_111623.json"], "model": "deepseek-ai/DeepSeek-V3.1-Base", "seed": 42, "lr": 1e-05, "steps": 30}
13
+ {"tinker_uri": "tinker://380ee7fe-a0fc-5224-b755-49256a020831:train:0/sampler_weights/step_50", "training_run_id": "380ee7fe-a0fc-5224-b755-49256a020831", "kind": "sampler_weights", "step": "step_50", "is_final": false, "referenced_in": ["atropos/notebooks/generate_notebooks.py"]}
14
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000100", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "sampler_weights", "step": 100, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
15
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000150", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "sampler_weights", "step": 150, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
16
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000200", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "sampler_weights", "step": 200, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
17
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000250", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "sampler_weights", "step": 250, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
18
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000300", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "sampler_weights", "step": 300, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
19
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000350", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "sampler_weights", "step": 350, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
20
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000400", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "sampler_weights", "step": 400, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
21
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000450", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "sampler_weights", "step": 450, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
22
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000050", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "sampler_weights", "step": 50, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
23
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000500", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "sampler_weights", "step": 500, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
24
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000550", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "sampler_weights", "step": 550, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
25
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000600", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "sampler_weights", "step": 600, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
26
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000650", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "sampler_weights", "step": 650, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
27
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000700", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "sampler_weights", "step": 700, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
28
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000100", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "weights", "step": 100, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
29
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000150", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "weights", "step": 150, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
30
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000200", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "weights", "step": 200, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
31
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000250", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "weights", "step": 250, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
32
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000300", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "weights", "step": 300, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
33
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000350", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "weights", "step": 350, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
34
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000400", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "weights", "step": 400, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
35
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000450", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "weights", "step": 450, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
36
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000050", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "weights", "step": 50, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
37
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000500", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "weights", "step": 500, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
38
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000550", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "weights", "step": 550, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
39
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000600", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "weights", "step": 600, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
40
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000650", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "weights", "step": 650, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
41
+ {"tinker_uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000700", "training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "kind": "weights", "step": 700, "is_final": false, "referenced_in": ["experiments/results/distillation_off_checkpoints.jsonl"]}
42
+ {"tinker_uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/sampler_weights/000100", "training_run_id": "39aa5eb2-e234-5a95-ab68-896e4cac8c45", "kind": "sampler_weights", "step": 100, "is_final": false, "referenced_in": ["experiments/results/arithmetic_checkpoints.jsonl"]}
43
+ {"tinker_uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/sampler_weights/000020", "training_run_id": "39aa5eb2-e234-5a95-ab68-896e4cac8c45", "kind": "sampler_weights", "step": 20, "is_final": false, "referenced_in": ["experiments/results/arithmetic_checkpoints.jsonl"]}
44
+ {"tinker_uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/sampler_weights/000040", "training_run_id": "39aa5eb2-e234-5a95-ab68-896e4cac8c45", "kind": "sampler_weights", "step": 40, "is_final": false, "referenced_in": ["experiments/results/arithmetic_checkpoints.jsonl"]}
45
+ {"tinker_uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/sampler_weights/000060", "training_run_id": "39aa5eb2-e234-5a95-ab68-896e4cac8c45", "kind": "sampler_weights", "step": 60, "is_final": false, "referenced_in": ["experiments/results/arithmetic_checkpoints.jsonl"]}
46
+ {"tinker_uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/sampler_weights/000080", "training_run_id": "39aa5eb2-e234-5a95-ab68-896e4cac8c45", "kind": "sampler_weights", "step": 80, "is_final": false, "referenced_in": ["experiments/results/arithmetic_checkpoints.jsonl"]}
47
+ {"tinker_uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/sampler_weights/final", "training_run_id": "39aa5eb2-e234-5a95-ab68-896e4cac8c45", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/results/arithmetic_checkpoints.jsonl"]}
48
+ {"tinker_uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/weights/000100", "training_run_id": "39aa5eb2-e234-5a95-ab68-896e4cac8c45", "kind": "weights", "step": 100, "is_final": false, "referenced_in": ["experiments/results/arithmetic_checkpoints.jsonl"]}
49
+ {"tinker_uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/weights/000020", "training_run_id": "39aa5eb2-e234-5a95-ab68-896e4cac8c45", "kind": "weights", "step": 20, "is_final": false, "referenced_in": ["experiments/results/arithmetic_checkpoints.jsonl"]}
50
+ {"tinker_uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/weights/000040", "training_run_id": "39aa5eb2-e234-5a95-ab68-896e4cac8c45", "kind": "weights", "step": 40, "is_final": false, "referenced_in": ["experiments/results/arithmetic_checkpoints.jsonl"]}
51
+ {"tinker_uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/weights/000060", "training_run_id": "39aa5eb2-e234-5a95-ab68-896e4cac8c45", "kind": "weights", "step": 60, "is_final": false, "referenced_in": ["experiments/results/arithmetic_checkpoints.jsonl"]}
52
+ {"tinker_uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/weights/000080", "training_run_id": "39aa5eb2-e234-5a95-ab68-896e4cac8c45", "kind": "weights", "step": 80, "is_final": false, "referenced_in": ["experiments/results/arithmetic_checkpoints.jsonl"]}
53
+ {"tinker_uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/weights/final", "training_run_id": "39aa5eb2-e234-5a95-ab68-896e4cac8c45", "kind": "weights", "step": "final", "is_final": true, "referenced_in": ["experiments/results/arithmetic_checkpoints.jsonl"]}
54
+ {"tinker_uri": "tinker://3fd600fd-6982-5061-8edc-c2da2c7f2f0e:train:0/sampler_weights/final", "training_run_id": "3fd600fd-6982-5061-8edc-c2da2c7f2f0e", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/results/heldout_gsm8k.json", "experiments/tinker-runs/results/campaign_v2_20260419_111623.json"], "model": "deepseek-ai/DeepSeek-V3.1", "seed": 789, "lr": 1e-05, "steps": 30}
55
+ {"tinker_uri": "tinker://5761c236-774f-5b29-926e-13bd6af310b1:train:0/sampler_weights/final", "training_run_id": "5761c236-774f-5b29-926e-13bd6af310b1", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/tinker-runs/results/campaign_v2_20260419_111623.json"], "model": "Qwen/Qwen3-8B", "seed": 42, "lr": 1e-05, "steps": 30}
56
+ {"tinker_uri": "tinker://57f4d068-2997-577c-895e-05f2265935a5:train:0/sampler_weights/final", "training_run_id": "57f4d068-2997-577c-895e-05f2265935a5", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/tinker-runs/results/campaign_v2_20260419_111623.json"], "model": "meta-llama/Llama-3.1-8B", "seed": 42, "lr": 1e-05, "steps": 30}
57
+ {"tinker_uri": "tinker://7040f210-e755-5f8d-bd1a-5cb6f3ab33f8:train:0/sampler_weights/final", "training_run_id": "7040f210-e755-5f8d-bd1a-5cb6f3ab33f8", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/results/heldout_gsm8k.json", "experiments/tinker-runs/results/campaign_v2_20260419_111623.json"], "model": "Qwen/Qwen3.5-397B-A17B", "seed": 42, "lr": 1e-05, "steps": 30}
58
+ {"tinker_uri": "tinker://9f27c001-b92c-55a7-9e12-1a8bd858e16d:train:0/sampler_weights/final", "training_run_id": "9f27c001-b92c-55a7-9e12-1a8bd858e16d", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/results/heldout_gsm8k.json", "experiments/tinker-runs/results/campaign_v2_20260419_111623.json"], "model": "Qwen/Qwen3-8B-Base", "seed": 42, "lr": 1e-05, "steps": 30}
59
+ {"tinker_uri": "tinker://a0d96a00-54e1-569c-a56c-98520819f14f:train:0/sampler_weights/final", "training_run_id": "a0d96a00-54e1-569c-a56c-98520819f14f", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/results/heldout_gsm8k.json", "experiments/tinker-runs/results/campaign_v2_20260419_111623.json"], "model": "openai/gpt-oss-120b", "seed": 42, "lr": 1e-05, "steps": 30}
60
+ {"tinker_uri": "tinker://a1d398a3-cea7-55fd-97d9-8b61001550b2:train:0/sampler_weights/final", "training_run_id": "a1d398a3-cea7-55fd-97d9-8b61001550b2", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/tinker-runs/results/campaign_v2_20260419_111623.json"], "model": "moonshotai/Kimi-K2.5", "seed": 42, "lr": 1e-05, "steps": 30}
61
+ {"tinker_uri": "tinker://a5715c67-65ef-5db8-a5ce-6d115a04b874:train:0/sampler_weights/final", "training_run_id": "a5715c67-65ef-5db8-a5ce-6d115a04b874", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/tinker-runs/results/campaign_v2_20260419_111623.json"], "model": "moonshotai/Kimi-K2-Thinking", "seed": 42, "lr": 1e-05, "steps": 30}
62
+ {"tinker_uri": "tinker://b39ea661-c094-5ecb-bed7-925c3e22f285:train:0/sampler_weights/final", "training_run_id": "b39ea661-c094-5ecb-bed7-925c3e22f285", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/tinker-runs/results/campaign_v2_20260419_111623.json"], "model": "meta-llama/Llama-3.1-70B", "seed": 42, "lr": 1e-05, "steps": 30}
63
+ {"tinker_uri": "tinker://c589e374-95d8-520d-8a4e-1b688bc5550f:train:0/sampler_weights/final", "training_run_id": "c589e374-95d8-520d-8a4e-1b688bc5550f", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/results/heldout_gsm8k.json", "experiments/tinker-runs/results/campaign_v2_20260419_111623.json"], "model": "deepseek-ai/DeepSeek-V3.1", "seed": 123, "lr": 1e-05, "steps": 30}
64
+ {"tinker_uri": "tinker://c82c876f-b1ff-517c-b19b-472b3df8e4b4:train:0/sampler_weights/final", "training_run_id": "c82c876f-b1ff-517c-b19b-472b3df8e4b4", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/tinker-runs/results/campaign_v2_20260419_111623.json"], "model": "Qwen/Qwen3-8B", "seed": 42, "lr": 1e-05, "steps": 30}
65
+ {"tinker_uri": "tinker://ca0d6a00-3a4c-5a24-8b8d-f9d69db9609e:train:0/sampler_weights/final", "training_run_id": "ca0d6a00-3a4c-5a24-8b8d-f9d69db9609e", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/results/heldout_gsm8k.json", "experiments/tinker-runs/results/campaign_v2_20260419_091120.json"], "model": "Qwen/Qwen3-8B-Base", "seed": 42, "lr": 1e-05, "steps": 30}
66
+ {"tinker_uri": "tinker://de39d965-b3f3-5317-b4e7-259a9472f2e0:train:0/sampler_weights/final", "training_run_id": "de39d965-b3f3-5317-b4e7-259a9472f2e0", "kind": "sampler_weights", "step": "final", "is_final": true, "referenced_in": ["experiments/tinker-runs/results/campaign_v2_20260419_111623.json"], "model": "Qwen/Qwen3-8B", "seed": 42, "lr": 1e-05, "steps": 30}
distillation_off_checkpoints.jsonl ADDED
@@ -0,0 +1,14 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"name": "000050", "epoch": 0, "batch": 50, "state_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000050", "sampler_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000050"}
2
+ {"name": "000100", "epoch": 0, "batch": 100, "state_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000100", "sampler_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000100"}
3
+ {"name": "000150", "epoch": 0, "batch": 150, "state_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000150", "sampler_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000150"}
4
+ {"name": "000200", "epoch": 0, "batch": 200, "state_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000200", "sampler_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000200"}
5
+ {"name": "000250", "epoch": 0, "batch": 250, "state_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000250", "sampler_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000250"}
6
+ {"name": "000300", "epoch": 0, "batch": 300, "state_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000300", "sampler_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000300"}
7
+ {"name": "000350", "epoch": 0, "batch": 350, "state_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000350", "sampler_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000350"}
8
+ {"name": "000400", "epoch": 0, "batch": 400, "state_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000400", "sampler_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000400"}
9
+ {"name": "000450", "epoch": 0, "batch": 450, "state_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000450", "sampler_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000450"}
10
+ {"name": "000500", "epoch": 0, "batch": 500, "state_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000500", "sampler_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000500"}
11
+ {"name": "000550", "epoch": 0, "batch": 550, "state_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000550", "sampler_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000550"}
12
+ {"name": "000600", "epoch": 0, "batch": 600, "state_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000600", "sampler_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000600"}
13
+ {"name": "000650", "epoch": 0, "batch": 650, "state_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000650", "sampler_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000650"}
14
+ {"name": "000700", "epoch": 0, "batch": 700, "state_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000700", "sampler_path": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000700"}
heldout_gsm8k.json ADDED
The diff for this file is too large to render. See raw diff
 
master_results.json ADDED
@@ -0,0 +1,4183 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metadata": {
3
+ "generated_at": "2026-04-19T02:40:19.496919Z",
4
+ "project": "tinker-rl-lab-world-class",
5
+ "wandb_project": "https://wandb.ai/arvindcr4-pes-university/tinker-rl-lab-world-class",
6
+ "total_experiments": 76,
7
+ "sources": [
8
+ "/home/user/workspace/tinker-rl-lab/experiments/all_results_consolidated.json",
9
+ "/home/user/workspace/tinker-rl-lab/experiments/tinker-runs/results",
10
+ "/home/user/workspace/tinker-rl-lab/experiments/modal/results/modal_parallel_results.json",
11
+ "/home/user/workspace/tinker-rl-lab/experiments/results/modal_results_all.json"
12
+ ],
13
+ "updated_at": "2026-04-19T11:17:17.107442+00:00"
14
+ },
15
+ "trl_grpo_baseline_summary": {
16
+ "mean_accuracy": 0.734,
17
+ "std_accuracy": 0.0703,
18
+ "seeds": [
19
+ 42,
20
+ 123,
21
+ 456,
22
+ 789,
23
+ 1024
24
+ ],
25
+ "accuracies": [
26
+ 0.735,
27
+ 0.81,
28
+ 0.62,
29
+ 0.74,
30
+ 0.765
31
+ ],
32
+ "model": "Qwen/Qwen2.5-0.5B",
33
+ "gpu": "NVIDIA L4",
34
+ "steps": 125
35
+ },
36
+ "experiments": [
37
+ {
38
+ "group": "Tinker GRPO",
39
+ "experiment_id": "frontier_gsm8k_deepseek-v3.1",
40
+ "experiment_name": "frontier gsm8k deepseek v3.1",
41
+ "model": "deepseek-ai/DeepSeek-V3.1",
42
+ "model_short": "deepseek-v3.1",
43
+ "method": "GRPO",
44
+ "task": "gsm8k",
45
+ "platform": "tinker",
46
+ "seed": 42,
47
+ "steps_completed": 20,
48
+ "peak_reward": 1.0,
49
+ "last10_avg": 0.85,
50
+ "first5_avg": 0.85,
51
+ "status": "completed",
52
+ "error_message": null,
53
+ "finding": null,
54
+ "wandb_run_url": "https://wandb.ai/arvindcr4-pes-university/tinker-rl-lab-world-class/runs/56b99b24-03bd-5d00-ae23-831933ef53b2",
55
+ "hf_checkpoint_url": "tinker://56b99b24-03bd-5d00-ae23-831933ef53b2:train:0/sampler_weights/final",
56
+ "reward_trace": [
57
+ 0.875,
58
+ 0.875,
59
+ 1.0,
60
+ 0.75,
61
+ 0.75,
62
+ 0.75,
63
+ 0.625,
64
+ 0.875,
65
+ 0.875,
66
+ 1.0,
67
+ 0.75,
68
+ 1.0,
69
+ 0.875,
70
+ 0.875,
71
+ 0.75,
72
+ 1.0,
73
+ 0.875,
74
+ 0.75,
75
+ 0.875,
76
+ 0.875
77
+ ],
78
+ "zero_reward_pct": 0.0,
79
+ "rank": 16,
80
+ "lr": 1e-05
81
+ },
82
+ {
83
+ "group": "Tinker GRPO",
84
+ "experiment_id": "cross_tool_llama-8b-inst",
85
+ "experiment_name": "cross tool llama 8b inst",
86
+ "model": "meta-llama/Llama-3.1-8B-Instruct",
87
+ "model_short": "llama-8b-inst",
88
+ "method": "GRPO",
89
+ "task": "tool_use",
90
+ "platform": "tinker",
91
+ "seed": 42,
92
+ "steps_completed": 30,
93
+ "peak_reward": 0.0,
94
+ "last10_avg": 0.0,
95
+ "first5_avg": 0.0,
96
+ "status": "failed",
97
+ "error_message": "Complete failure - tool_use task too difficult for this model",
98
+ "finding": "Complete failure \u2014 tool_use task too hard for Llama-3.1-8B-Instruct (100% zero reward)",
99
+ "wandb_run_url": "https://wandb.ai/arvindcr4-pes-university/tinker-rl-lab-world-class/runs/ca2e3a24-7401-5770-af34-a0d27177aeaa",
100
+ "hf_checkpoint_url": "tinker://ca2e3a24-7401-5770-af34-a0d27177aeaa:train:0/sampler_weights/final",
101
+ "reward_trace": [
102
+ 0.0,
103
+ 0.0,
104
+ 0.0,
105
+ 0.0,
106
+ 0.0,
107
+ 0.0,
108
+ 0.0,
109
+ 0.0,
110
+ 0.0,
111
+ 0.0,
112
+ 0.0,
113
+ 0.0,
114
+ 0.0,
115
+ 0.0,
116
+ 0.0,
117
+ 0.0,
118
+ 0.0,
119
+ 0.0,
120
+ 0.0,
121
+ 0.0,
122
+ 0.0,
123
+ 0.0,
124
+ 0.0,
125
+ 0.0,
126
+ 0.0,
127
+ 0.0,
128
+ 0.0,
129
+ 0.0,
130
+ 0.0,
131
+ 0.0
132
+ ],
133
+ "zero_reward_pct": 100.0,
134
+ "rank": 32,
135
+ "lr": 3e-05
136
+ },
137
+ {
138
+ "group": "Tinker GRPO",
139
+ "experiment_id": "scale_gsm8k_qwen3-8b",
140
+ "experiment_name": "scale gsm8k qwen3 8b",
141
+ "model": "Qwen/Qwen3-8B",
142
+ "model_short": "qwen3-8b",
143
+ "method": "GRPO",
144
+ "task": "gsm8k",
145
+ "platform": "tinker",
146
+ "seed": 42,
147
+ "steps_completed": 30,
148
+ "peak_reward": 0.625,
149
+ "last10_avg": 0.34375,
150
+ "first5_avg": 0.3125,
151
+ "status": "completed",
152
+ "error_message": null,
153
+ "finding": null,
154
+ "wandb_run_url": "https://wandb.ai/arvindcr4-pes-university/tinker-rl-lab-world-class/runs/3154dee5-4fd7-59b6-ba3b-721eef675bfc",
155
+ "hf_checkpoint_url": "tinker://3154dee5-4fd7-59b6-ba3b-721eef675bfc:train:0/sampler_weights/final",
156
+ "reward_trace": [
157
+ 0.375,
158
+ 0.0625,
159
+ 0.5625,
160
+ 0.1875,
161
+ 0.375,
162
+ 0.3125,
163
+ 0.125,
164
+ 0.125,
165
+ 0.1875,
166
+ 0.125,
167
+ 0.25,
168
+ 0.4375,
169
+ 0.625,
170
+ 0.1875,
171
+ 0.3125,
172
+ 0.5,
173
+ 0.375,
174
+ 0.375,
175
+ 0.4375,
176
+ 0.5,
177
+ 0.3125,
178
+ 0.125,
179
+ 0.25,
180
+ 0.4375,
181
+ 0.5,
182
+ 0.3125,
183
+ 0.4375,
184
+ 0.375,
185
+ 0.1875,
186
+ 0.25
187
+ ],
188
+ "zero_reward_pct": 6.666666666666667,
189
+ "rank": 32,
190
+ "lr": 3e-05
191
+ },
192
+ {
193
+ "group": "Tinker GRPO",
194
+ "experiment_id": "scale_gsm8k_qwen3-32b",
195
+ "experiment_name": "scale gsm8k qwen3 32b",
196
+ "model": "qwen3-32b",
197
+ "model_short": "qwen3-32b",
198
+ "method": "GRPO",
199
+ "task": "gsm8k",
200
+ "platform": "tinker",
201
+ "seed": 42,
202
+ "steps_completed": 0,
203
+ "peak_reward": null,
204
+ "last10_avg": null,
205
+ "first5_avg": null,
206
+ "status": "failed",
207
+ "error_message": "JWT authentication failure \u2014 run did not start",
208
+ "finding": "Blocked by Tinker JWT auth error; no training data collected",
209
+ "wandb_run_url": null,
210
+ "hf_checkpoint_url": null,
211
+ "reward_trace": [],
212
+ "zero_reward_pct": null,
213
+ "rank": null,
214
+ "lr": null
215
+ },
216
+ {
217
+ "group": "Tinker GRPO",
218
+ "experiment_id": "scale_gsm8k_qwen3.5-4b",
219
+ "experiment_name": "scale gsm8k qwen3.5 4b",
220
+ "model": "qwen3.5-4b",
221
+ "model_short": "qwen3.5-4b",
222
+ "method": "GRPO",
223
+ "task": "gsm8k",
224
+ "platform": "tinker",
225
+ "seed": 42,
226
+ "steps_completed": 0,
227
+ "peak_reward": null,
228
+ "last10_avg": null,
229
+ "first5_avg": null,
230
+ "status": "failed",
231
+ "error_message": "JWT authentication failure \u2014 run did not start",
232
+ "finding": "Blocked by Tinker JWT auth error; no training data collected",
233
+ "wandb_run_url": null,
234
+ "hf_checkpoint_url": null,
235
+ "reward_trace": [],
236
+ "zero_reward_pct": null,
237
+ "rank": null,
238
+ "lr": null
239
+ },
240
+ {
241
+ "group": "Tinker GRPO",
242
+ "experiment_id": "scale_gsm8k_qwen3.5-27b",
243
+ "experiment_name": "scale gsm8k qwen3.5 27b",
244
+ "model": "qwen3.5-27b",
245
+ "model_short": "qwen3.5-27b",
246
+ "method": "GRPO",
247
+ "task": "gsm8k",
248
+ "platform": "tinker",
249
+ "seed": 42,
250
+ "steps_completed": 0,
251
+ "peak_reward": null,
252
+ "last10_avg": null,
253
+ "first5_avg": null,
254
+ "status": "failed",
255
+ "error_message": "JWT authentication failure \u2014 run did not start",
256
+ "finding": "Blocked by Tinker JWT auth error; no training data collected",
257
+ "wandb_run_url": null,
258
+ "hf_checkpoint_url": null,
259
+ "reward_trace": [],
260
+ "zero_reward_pct": null,
261
+ "rank": null,
262
+ "lr": null
263
+ },
264
+ {
265
+ "group": "Tinker GRPO",
266
+ "experiment_id": "scale_gsm8k_llama-8b-inst",
267
+ "experiment_name": "scale gsm8k llama 8b inst",
268
+ "model": "llama-8b-inst",
269
+ "model_short": "llama-8b-inst",
270
+ "method": "GRPO",
271
+ "task": "gsm8k",
272
+ "platform": "tinker",
273
+ "seed": 42,
274
+ "steps_completed": 0,
275
+ "peak_reward": null,
276
+ "last10_avg": null,
277
+ "first5_avg": null,
278
+ "status": "failed",
279
+ "error_message": "JWT authentication failure \u2014 run did not start",
280
+ "finding": "Blocked by Tinker JWT auth error; no training data collected",
281
+ "wandb_run_url": null,
282
+ "hf_checkpoint_url": null,
283
+ "reward_trace": [],
284
+ "zero_reward_pct": null,
285
+ "rank": null,
286
+ "lr": null
287
+ },
288
+ {
289
+ "group": "Tinker GRPO",
290
+ "experiment_id": "frontier_gsm8k_qwen3-235b",
291
+ "experiment_name": "frontier gsm8k qwen3 235b",
292
+ "model": "qwen3-235b",
293
+ "model_short": "qwen3-235b",
294
+ "method": "GRPO",
295
+ "task": "gsm8k",
296
+ "platform": "tinker",
297
+ "seed": 42,
298
+ "steps_completed": 0,
299
+ "peak_reward": null,
300
+ "last10_avg": null,
301
+ "first5_avg": null,
302
+ "status": "failed",
303
+ "error_message": "JWT authentication failure \u2014 run did not start",
304
+ "finding": "Blocked by Tinker JWT auth error; no training data collected",
305
+ "wandb_run_url": null,
306
+ "hf_checkpoint_url": null,
307
+ "reward_trace": [],
308
+ "zero_reward_pct": null,
309
+ "rank": null,
310
+ "lr": null
311
+ },
312
+ {
313
+ "group": "Tinker GRPO",
314
+ "experiment_id": "frontier_gsm8k_nemotron-120b",
315
+ "experiment_name": "frontier gsm8k nemotron 120b",
316
+ "model": "nemotron-120b",
317
+ "model_short": "nemotron-120b",
318
+ "method": "GRPO",
319
+ "task": "gsm8k",
320
+ "platform": "tinker",
321
+ "seed": 42,
322
+ "steps_completed": 0,
323
+ "peak_reward": null,
324
+ "last10_avg": null,
325
+ "first5_avg": null,
326
+ "status": "failed",
327
+ "error_message": "JWT authentication failure \u2014 run did not start",
328
+ "finding": "Blocked by Tinker JWT auth error; no training data collected",
329
+ "wandb_run_url": null,
330
+ "hf_checkpoint_url": null,
331
+ "reward_trace": [],
332
+ "zero_reward_pct": null,
333
+ "rank": null,
334
+ "lr": null
335
+ },
336
+ {
337
+ "group": "Tinker GRPO",
338
+ "experiment_id": "moe_gsm8k_qwen3-30b-moe",
339
+ "experiment_name": "moe gsm8k qwen3 30b moe",
340
+ "model": "qwen3-30b-moe",
341
+ "model_short": "qwen3-30b-moe",
342
+ "method": "GRPO",
343
+ "task": "gsm8k",
344
+ "platform": "tinker",
345
+ "seed": 42,
346
+ "steps_completed": 0,
347
+ "peak_reward": null,
348
+ "last10_avg": null,
349
+ "first5_avg": null,
350
+ "status": "failed",
351
+ "error_message": "JWT authentication failure \u2014 run did not start",
352
+ "finding": "Blocked by Tinker JWT auth error; no training data collected",
353
+ "wandb_run_url": null,
354
+ "hf_checkpoint_url": null,
355
+ "reward_trace": [],
356
+ "zero_reward_pct": null,
357
+ "rank": null,
358
+ "lr": null
359
+ },
360
+ {
361
+ "group": "Tinker GRPO",
362
+ "experiment_id": "moe_gsm8k_qwen3-30b-inst",
363
+ "experiment_name": "moe gsm8k qwen3 30b inst",
364
+ "model": "qwen3-30b-inst",
365
+ "model_short": "qwen3-30b-inst",
366
+ "method": "GRPO",
367
+ "task": "gsm8k",
368
+ "platform": "tinker",
369
+ "seed": 42,
370
+ "steps_completed": 0,
371
+ "peak_reward": null,
372
+ "last10_avg": null,
373
+ "first5_avg": null,
374
+ "status": "failed",
375
+ "error_message": "JWT authentication failure \u2014 run did not start",
376
+ "finding": "Blocked by Tinker JWT auth error; no training data collected",
377
+ "wandb_run_url": null,
378
+ "hf_checkpoint_url": null,
379
+ "reward_trace": [],
380
+ "zero_reward_pct": null,
381
+ "rank": null,
382
+ "lr": null
383
+ },
384
+ {
385
+ "group": "Tinker GRPO",
386
+ "experiment_id": "cross_tool_qwen3-32b",
387
+ "experiment_name": "cross tool qwen3 32b",
388
+ "model": "qwen3-32b",
389
+ "model_short": "qwen3-32b",
390
+ "method": "GRPO",
391
+ "task": "tool",
392
+ "platform": "tinker",
393
+ "seed": 42,
394
+ "steps_completed": 0,
395
+ "peak_reward": null,
396
+ "last10_avg": null,
397
+ "first5_avg": null,
398
+ "status": "failed",
399
+ "error_message": "JWT authentication failure \u2014 run did not start",
400
+ "finding": "Blocked by Tinker JWT auth error; no training data collected",
401
+ "wandb_run_url": null,
402
+ "hf_checkpoint_url": null,
403
+ "reward_trace": [],
404
+ "zero_reward_pct": null,
405
+ "rank": null,
406
+ "lr": null
407
+ },
408
+ {
409
+ "group": "Tinker GRPO",
410
+ "experiment_id": "arch_gsm8k_gpt-oss-20b",
411
+ "experiment_name": "arch gsm8k gpt oss 20b",
412
+ "model": "gpt-oss-20b",
413
+ "model_short": "gpt-oss-20b",
414
+ "method": "GRPO",
415
+ "task": "gsm8k",
416
+ "platform": "tinker",
417
+ "seed": 42,
418
+ "steps_completed": 0,
419
+ "peak_reward": null,
420
+ "last10_avg": null,
421
+ "first5_avg": null,
422
+ "status": "failed",
423
+ "error_message": "JWT authentication failure \u2014 run did not start",
424
+ "finding": "Blocked by Tinker JWT auth error; no training data collected",
425
+ "wandb_run_url": null,
426
+ "hf_checkpoint_url": null,
427
+ "reward_trace": [],
428
+ "zero_reward_pct": null,
429
+ "rank": null,
430
+ "lr": null
431
+ },
432
+ {
433
+ "group": "Tinker GRPO",
434
+ "experiment_id": "arch_gsm8k_kimi-k2",
435
+ "experiment_name": "arch gsm8k kimi k2",
436
+ "model": "kimi-k2",
437
+ "model_short": "kimi-k2",
438
+ "method": "GRPO",
439
+ "task": "gsm8k",
440
+ "platform": "tinker",
441
+ "seed": 42,
442
+ "steps_completed": 0,
443
+ "peak_reward": null,
444
+ "last10_avg": null,
445
+ "first5_avg": null,
446
+ "status": "failed",
447
+ "error_message": "JWT authentication failure \u2014 run did not start",
448
+ "finding": "Blocked by Tinker JWT auth error; no training data collected",
449
+ "wandb_run_url": null,
450
+ "hf_checkpoint_url": null,
451
+ "reward_trace": [],
452
+ "zero_reward_pct": null,
453
+ "rank": null,
454
+ "lr": null
455
+ },
456
+ {
457
+ "group": "Modal PPO",
458
+ "experiment_id": "ppo_qwen3-8b",
459
+ "experiment_name": "ppo_gsm8k_Qwen3-8B_s42",
460
+ "model": "Qwen/Qwen3-8B",
461
+ "model_short": "qwen3-8b",
462
+ "method": "PPO-REINFORCE",
463
+ "task": "gsm8k",
464
+ "platform": "modal_h100",
465
+ "seed": 42,
466
+ "steps_completed": 30,
467
+ "peak_reward": 1.0,
468
+ "last10_avg": 0.35,
469
+ "first5_avg": 0.4,
470
+ "status": "completed",
471
+ "error_message": null,
472
+ "finding": "High variance \u2014 Qwen3-8B peaks 1.0 but last-10 avg only 0.35; unstable PPO training",
473
+ "wandb_run_url": "https://wandb.ai/arvindcr4-pes-university/tinker-rl-lab-world-class/runs/ppo-gsm8k-Qwen3-8B-s42",
474
+ "hf_checkpoint_url": "https://huggingface.co/arvindcr4/tinker-rl-bench-ppo_gsm8k_Qwen3-8B_s42",
475
+ "reward_trace": [
476
+ 0.5,
477
+ 0.5,
478
+ 0.0,
479
+ 0.0,
480
+ 1.0,
481
+ 0.5,
482
+ 0.0,
483
+ 0.0,
484
+ 0.25,
485
+ 0.5,
486
+ 0.5,
487
+ 0.0,
488
+ 0.5,
489
+ 0.25,
490
+ 0.0,
491
+ 0.0,
492
+ 0.0,
493
+ 0.25,
494
+ 0.25,
495
+ 0.0,
496
+ 0.5,
497
+ 0.5,
498
+ 0.5,
499
+ 0.0,
500
+ 0.75,
501
+ 0.25,
502
+ 0.0,
503
+ 0.25,
504
+ 0.75,
505
+ 0.0
506
+ ],
507
+ "zero_reward_pct": null,
508
+ "rank": null,
509
+ "lr": null
510
+ },
511
+ {
512
+ "group": "Modal PPO",
513
+ "experiment_id": "ppo_llama-8b-inst",
514
+ "experiment_name": "ppo_gsm8k_Llama-3.1-8B-Instruct_s42",
515
+ "model": "meta-llama/Llama-3.1-8B-Instruct",
516
+ "model_short": "llama-8b-inst",
517
+ "method": "PPO-REINFORCE",
518
+ "task": "gsm8k",
519
+ "platform": "modal_h100",
520
+ "seed": 42,
521
+ "steps_completed": 30,
522
+ "peak_reward": 1.0,
523
+ "last10_avg": 0.95,
524
+ "first5_avg": 0.95,
525
+ "status": "completed",
526
+ "error_message": null,
527
+ "finding": "Strong performance \u2014 Llama-3.1-8B peaks 1.0, last-10 avg 0.95 on GSM8K",
528
+ "wandb_run_url": "https://wandb.ai/arvindcr4-pes-university/tinker-rl-lab-world-class/runs/ppo-gsm8k-Llama-3.1-8B-Instruct-s42",
529
+ "hf_checkpoint_url": null,
530
+ "reward_trace": [
531
+ 1.0,
532
+ 1.0,
533
+ 1.0,
534
+ 0.75,
535
+ 1.0,
536
+ 1.0,
537
+ 1.0,
538
+ 1.0,
539
+ 1.0,
540
+ 1.0,
541
+ 1.0,
542
+ 0.5,
543
+ 1.0,
544
+ 1.0,
545
+ 1.0,
546
+ 1.0,
547
+ 1.0,
548
+ 1.0,
549
+ 0.75,
550
+ 1.0,
551
+ 1.0,
552
+ 1.0,
553
+ 1.0,
554
+ 0.75,
555
+ 1.0,
556
+ 1.0,
557
+ 1.0,
558
+ 1.0,
559
+ 0.75,
560
+ 1.0
561
+ ],
562
+ "zero_reward_pct": null,
563
+ "rank": null,
564
+ "lr": null
565
+ },
566
+ {
567
+ "group": "Modal Other",
568
+ "experiment_id": "humaneval_qwen3-8b",
569
+ "experiment_name": "humaneval qwen3 8b",
570
+ "model": "qwen3-8b",
571
+ "model_short": "qwen3-8b",
572
+ "method": "eval",
573
+ "task": "humaneval",
574
+ "platform": "modal_h100",
575
+ "seed": 42,
576
+ "steps_completed": null,
577
+ "peak_reward": null,
578
+ "last10_avg": null,
579
+ "first5_avg": null,
580
+ "status": "partial",
581
+ "error_message": "timeout 3600s",
582
+ "finding": "Partial result: 40/164 problems, cum_pass@1=65.0%; timed out at 3600s",
583
+ "wandb_run_url": null,
584
+ "hf_checkpoint_url": null,
585
+ "reward_trace": [],
586
+ "zero_reward_pct": null,
587
+ "rank": null,
588
+ "lr": null
589
+ },
590
+ {
591
+ "group": "Modal Other",
592
+ "experiment_id": "heldout_qwen3.5-27b",
593
+ "experiment_name": "heldout qwen3.5 27b",
594
+ "model": "qwen3.5-27b",
595
+ "model_short": "qwen3.5-27b",
596
+ "method": "eval",
597
+ "task": "heldout",
598
+ "platform": "modal_h100",
599
+ "seed": 42,
600
+ "steps_completed": null,
601
+ "peak_reward": null,
602
+ "last10_avg": null,
603
+ "first5_avg": null,
604
+ "status": "partial",
605
+ "error_message": "timeout 3600s",
606
+ "finding": "Partial result: 100/200 samples, 86/100=86.0%; timed out at 3600s",
607
+ "wandb_run_url": null,
608
+ "hf_checkpoint_url": null,
609
+ "reward_trace": [],
610
+ "zero_reward_pct": null,
611
+ "rank": null,
612
+ "lr": null
613
+ },
614
+ {
615
+ "group": "Modal Other",
616
+ "experiment_id": "heldout_qwen3-32b",
617
+ "experiment_name": "heldout qwen3 32b",
618
+ "model": "qwen3-32b",
619
+ "model_short": "qwen3-32b",
620
+ "method": "eval",
621
+ "task": "heldout",
622
+ "platform": "modal_h100",
623
+ "seed": 42,
624
+ "steps_completed": null,
625
+ "peak_reward": null,
626
+ "last10_avg": null,
627
+ "first5_avg": null,
628
+ "status": "partial",
629
+ "error_message": "timeout 3600s",
630
+ "finding": "Partial result: 100/200 samples, 33/100=33.0%; timed out at 3600s",
631
+ "wandb_run_url": null,
632
+ "hf_checkpoint_url": null,
633
+ "reward_trace": [],
634
+ "zero_reward_pct": null,
635
+ "rank": null,
636
+ "lr": null
637
+ },
638
+ {
639
+ "group": "Modal Other",
640
+ "experiment_id": "kl_qwen3-8b",
641
+ "experiment_name": "kl qwen3 8b",
642
+ "model": "Qwen3-8B",
643
+ "model_short": "qwen3-8b",
644
+ "method": "PPO+KL",
645
+ "task": "gsm8k",
646
+ "platform": "modal_h100",
647
+ "seed": 42,
648
+ "steps_completed": 0,
649
+ "peak_reward": null,
650
+ "last10_avg": null,
651
+ "first5_avg": null,
652
+ "status": "failed",
653
+ "error_message": "element 0 of tensors does not require grad and does not have a grad_fn",
654
+ "finding": "Gradient error \u2014 KL divergence PPO variant failed immediately on Qwen3-8B",
655
+ "wandb_run_url": null,
656
+ "hf_checkpoint_url": null,
657
+ "reward_trace": [],
658
+ "zero_reward_pct": null,
659
+ "rank": null,
660
+ "lr": null
661
+ },
662
+ {
663
+ "group": "Old TRL",
664
+ "experiment_id": "trl_grpo_math_s42",
665
+ "experiment_name": "TRL GRPO Math seed=42",
666
+ "model": "Qwen/Qwen2.5-0.5B",
667
+ "model_short": "qwen2.5-0.5b",
668
+ "method": "GRPO",
669
+ "task": "math",
670
+ "platform": "modal_nvidial4",
671
+ "seed": 42,
672
+ "steps_completed": 125,
673
+ "peak_reward": 0.735,
674
+ "last10_avg": 0.735,
675
+ "first5_avg": null,
676
+ "status": "completed",
677
+ "error_message": null,
678
+ "finding": "Seed 42: accuracy=0.735, loss=0.00896, elapsed=141s",
679
+ "wandb_run_url": null,
680
+ "hf_checkpoint_url": null,
681
+ "reward_trace": [],
682
+ "zero_reward_pct": null,
683
+ "rank": null,
684
+ "lr": null,
685
+ "train_loss": 0.00895642852783203,
686
+ "elapsed_seconds": 140.5099287033081
687
+ },
688
+ {
689
+ "group": "Old TRL",
690
+ "experiment_id": "trl_grpo_math_s123",
691
+ "experiment_name": "TRL GRPO Math seed=123",
692
+ "model": "Qwen/Qwen2.5-0.5B",
693
+ "model_short": "qwen2.5-0.5b",
694
+ "method": "GRPO",
695
+ "task": "math",
696
+ "platform": "modal_nvidial4",
697
+ "seed": 123,
698
+ "steps_completed": 125,
699
+ "peak_reward": 0.81,
700
+ "last10_avg": 0.81,
701
+ "first5_avg": null,
702
+ "status": "completed",
703
+ "error_message": null,
704
+ "finding": "Seed 123: accuracy=0.810, loss=0.00699, elapsed=135s",
705
+ "wandb_run_url": null,
706
+ "hf_checkpoint_url": null,
707
+ "reward_trace": [],
708
+ "zero_reward_pct": null,
709
+ "rank": null,
710
+ "lr": null,
711
+ "train_loss": 0.006986120015382766,
712
+ "elapsed_seconds": 135.4678463935852
713
+ },
714
+ {
715
+ "group": "Old TRL",
716
+ "experiment_id": "trl_grpo_math_s456",
717
+ "experiment_name": "TRL GRPO Math seed=456",
718
+ "model": "Qwen/Qwen2.5-0.5B",
719
+ "model_short": "qwen2.5-0.5b",
720
+ "method": "GRPO",
721
+ "task": "math",
722
+ "platform": "modal_nvidial4",
723
+ "seed": 456,
724
+ "steps_completed": 125,
725
+ "peak_reward": 0.62,
726
+ "last10_avg": 0.62,
727
+ "first5_avg": null,
728
+ "status": "completed",
729
+ "error_message": null,
730
+ "finding": "Seed 456: accuracy=0.620, loss=0.00792, elapsed=142s",
731
+ "wandb_run_url": null,
732
+ "hf_checkpoint_url": null,
733
+ "reward_trace": [],
734
+ "zero_reward_pct": null,
735
+ "rank": null,
736
+ "lr": null,
737
+ "train_loss": 0.007916679546236993,
738
+ "elapsed_seconds": 142.4041965007782
739
+ },
740
+ {
741
+ "group": "Old TRL",
742
+ "experiment_id": "trl_grpo_math_s789",
743
+ "experiment_name": "TRL GRPO Math seed=789",
744
+ "model": "Qwen/Qwen2.5-0.5B",
745
+ "model_short": "qwen2.5-0.5b",
746
+ "method": "GRPO",
747
+ "task": "math",
748
+ "platform": "modal_nvidial4",
749
+ "seed": 789,
750
+ "steps_completed": 125,
751
+ "peak_reward": 0.74,
752
+ "last10_avg": 0.74,
753
+ "first5_avg": null,
754
+ "status": "completed",
755
+ "error_message": null,
756
+ "finding": "Seed 789: accuracy=0.740, loss=0.00322, elapsed=141s",
757
+ "wandb_run_url": null,
758
+ "hf_checkpoint_url": null,
759
+ "reward_trace": [],
760
+ "zero_reward_pct": null,
761
+ "rank": null,
762
+ "lr": null,
763
+ "train_loss": 0.003221572862006724,
764
+ "elapsed_seconds": 140.53626608848572
765
+ },
766
+ {
767
+ "group": "Old TRL",
768
+ "experiment_id": "trl_grpo_math_s1024",
769
+ "experiment_name": "TRL GRPO Math seed=1024",
770
+ "model": "Qwen/Qwen2.5-0.5B",
771
+ "model_short": "qwen2.5-0.5b",
772
+ "method": "GRPO",
773
+ "task": "math",
774
+ "platform": "modal_nvidial4",
775
+ "seed": 1024,
776
+ "steps_completed": 125,
777
+ "peak_reward": 0.765,
778
+ "last10_avg": 0.765,
779
+ "first5_avg": null,
780
+ "status": "completed",
781
+ "error_message": null,
782
+ "finding": "Seed 1024: accuracy=0.765, loss=0.00418, elapsed=183s",
783
+ "wandb_run_url": null,
784
+ "hf_checkpoint_url": null,
785
+ "reward_trace": [],
786
+ "zero_reward_pct": null,
787
+ "rank": null,
788
+ "lr": null,
789
+ "train_loss": 0.004181556344032287,
790
+ "elapsed_seconds": 182.66980004310608
791
+ },
792
+ {
793
+ "group": "Old TRL",
794
+ "experiment_id": "sb3_ppo_math_s42",
795
+ "experiment_name": "SB3 PPO Math seed=42",
796
+ "model": "SB3/PPO (policy network)",
797
+ "model_short": "sb3-ppo",
798
+ "method": "PPO",
799
+ "task": "math",
800
+ "platform": "modal",
801
+ "seed": 42,
802
+ "steps_completed": 100352,
803
+ "peak_reward": 0.035,
804
+ "last10_avg": 0.008,
805
+ "first5_avg": 0.009,
806
+ "status": "completed",
807
+ "error_message": null,
808
+ "finding": "SB3/PPO: final accuracy=0.003 \u2014 policy gradient on raw math fails without LLM",
809
+ "wandb_run_url": null,
810
+ "hf_checkpoint_url": null,
811
+ "reward_trace": [
812
+ 0.005,
813
+ 0.02,
814
+ 0.005,
815
+ 0.01,
816
+ 0.005,
817
+ 0.01,
818
+ 0.01,
819
+ 0.0,
820
+ 0.0,
821
+ 0.0,
822
+ 0.01,
823
+ 0.0,
824
+ 0.005,
825
+ 0.005,
826
+ 0.005,
827
+ 0.0,
828
+ 0.005,
829
+ 0.0,
830
+ 0.0,
831
+ 0.02,
832
+ 0.005,
833
+ 0.01,
834
+ 0.01,
835
+ 0.015,
836
+ 0.01,
837
+ 0.025,
838
+ 0.005,
839
+ 0.005,
840
+ 0.02,
841
+ 0.03,
842
+ 0.0,
843
+ 0.005,
844
+ 0.035,
845
+ 0.005,
846
+ 0.005,
847
+ 0.0,
848
+ 0.0,
849
+ 0.005,
850
+ 0.01,
851
+ 0.005,
852
+ 0.015,
853
+ 0.01,
854
+ 0.005,
855
+ 0.005,
856
+ 0.005,
857
+ 0.01,
858
+ 0.0,
859
+ 0.01,
860
+ 0.015
861
+ ],
862
+ "elapsed_seconds": 189.52233171463013
863
+ },
864
+ {
865
+ "group": "Old TRL",
866
+ "experiment_id": "sb3_ppo_math_s123",
867
+ "experiment_name": "SB3 PPO Math seed=123",
868
+ "model": "SB3/PPO (policy network)",
869
+ "model_short": "sb3-ppo",
870
+ "method": "PPO",
871
+ "task": "math",
872
+ "platform": "modal",
873
+ "seed": 123,
874
+ "steps_completed": 100352,
875
+ "peak_reward": 0.045,
876
+ "last10_avg": 0.008,
877
+ "first5_avg": 0.008,
878
+ "status": "completed",
879
+ "error_message": null,
880
+ "finding": "SB3/PPO: final accuracy=0.014 \u2014 policy gradient on raw math fails without LLM",
881
+ "wandb_run_url": null,
882
+ "hf_checkpoint_url": null,
883
+ "reward_trace": [
884
+ 0.01,
885
+ 0.0,
886
+ 0.015,
887
+ 0.0,
888
+ 0.015,
889
+ 0.005,
890
+ 0.015,
891
+ 0.01,
892
+ 0.015,
893
+ 0.025,
894
+ 0.015,
895
+ 0.01,
896
+ 0.005,
897
+ 0.01,
898
+ 0.01,
899
+ 0.015,
900
+ 0.01,
901
+ 0.015,
902
+ 0.015,
903
+ 0.0,
904
+ 0.015,
905
+ 0.02,
906
+ 0.005,
907
+ 0.005,
908
+ 0.005,
909
+ 0.015,
910
+ 0.015,
911
+ 0.015,
912
+ 0.005,
913
+ 0.045,
914
+ 0.005,
915
+ 0.01,
916
+ 0.0,
917
+ 0.005,
918
+ 0.01,
919
+ 0.015,
920
+ 0.005,
921
+ 0.0,
922
+ 0.0,
923
+ 0.015,
924
+ 0.005,
925
+ 0.01,
926
+ 0.005,
927
+ 0.005,
928
+ 0.0,
929
+ 0.01,
930
+ 0.015,
931
+ 0.0,
932
+ 0.015
933
+ ],
934
+ "elapsed_seconds": 187.3643729686737
935
+ },
936
+ {
937
+ "group": "Old TRL",
938
+ "experiment_id": "sb3_ppo_math_s456",
939
+ "experiment_name": "SB3 PPO Math seed=456",
940
+ "model": "SB3/PPO (policy network)",
941
+ "model_short": "sb3-ppo",
942
+ "method": "PPO",
943
+ "task": "math",
944
+ "platform": "modal",
945
+ "seed": 456,
946
+ "steps_completed": 100352,
947
+ "peak_reward": 0.02,
948
+ "last10_avg": 0.0065,
949
+ "first5_avg": 0.007,
950
+ "status": "completed",
951
+ "error_message": null,
952
+ "finding": "SB3/PPO: final accuracy=0.011 \u2014 policy gradient on raw math fails without LLM",
953
+ "wandb_run_url": null,
954
+ "hf_checkpoint_url": null,
955
+ "reward_trace": [
956
+ 0.005,
957
+ 0.005,
958
+ 0.01,
959
+ 0.01,
960
+ 0.005,
961
+ 0.01,
962
+ 0.005,
963
+ 0.0,
964
+ 0.02,
965
+ 0.01,
966
+ 0.005,
967
+ 0.015,
968
+ 0.005,
969
+ 0.0,
970
+ 0.02,
971
+ 0.005,
972
+ 0.01,
973
+ 0.01,
974
+ 0.01,
975
+ 0.005,
976
+ 0.005,
977
+ 0.005,
978
+ 0.0,
979
+ 0.015,
980
+ 0.015,
981
+ 0.015,
982
+ 0.02,
983
+ 0.01,
984
+ 0.015,
985
+ 0.005,
986
+ 0.0,
987
+ 0.0,
988
+ 0.0,
989
+ 0.005,
990
+ 0.02,
991
+ 0.005,
992
+ 0.005,
993
+ 0.0,
994
+ 0.01,
995
+ 0.0,
996
+ 0.0,
997
+ 0.0,
998
+ 0.005,
999
+ 0.02,
1000
+ 0.0,
1001
+ 0.01,
1002
+ 0.015,
1003
+ 0.0,
1004
+ 0.015
1005
+ ],
1006
+ "elapsed_seconds": 192.28725409507751
1007
+ },
1008
+ {
1009
+ "group": "Old TRL",
1010
+ "experiment_id": "sb3_ppo_math_s789",
1011
+ "experiment_name": "SB3 PPO Math seed=789",
1012
+ "model": "SB3/PPO (policy network)",
1013
+ "model_short": "sb3-ppo",
1014
+ "method": "PPO",
1015
+ "task": "math",
1016
+ "platform": "modal",
1017
+ "seed": 789,
1018
+ "steps_completed": 100352,
1019
+ "peak_reward": 0.02,
1020
+ "last10_avg": 0.01,
1021
+ "first5_avg": 0.003,
1022
+ "status": "completed",
1023
+ "error_message": null,
1024
+ "finding": "SB3/PPO: final accuracy=0.012 \u2014 policy gradient on raw math fails without LLM",
1025
+ "wandb_run_url": null,
1026
+ "hf_checkpoint_url": null,
1027
+ "reward_trace": [
1028
+ 0.0,
1029
+ 0.01,
1030
+ 0.005,
1031
+ 0.0,
1032
+ 0.0,
1033
+ 0.005,
1034
+ 0.005,
1035
+ 0.015,
1036
+ 0.02,
1037
+ 0.005,
1038
+ 0.01,
1039
+ 0.005,
1040
+ 0.005,
1041
+ 0.0,
1042
+ 0.005,
1043
+ 0.01,
1044
+ 0.015,
1045
+ 0.0,
1046
+ 0.005,
1047
+ 0.01,
1048
+ 0.0,
1049
+ 0.0,
1050
+ 0.005,
1051
+ 0.0,
1052
+ 0.0,
1053
+ 0.0,
1054
+ 0.01,
1055
+ 0.005,
1056
+ 0.015,
1057
+ 0.01,
1058
+ 0.015,
1059
+ 0.0,
1060
+ 0.0,
1061
+ 0.0,
1062
+ 0.0,
1063
+ 0.015,
1064
+ 0.01,
1065
+ 0.015,
1066
+ 0.0,
1067
+ 0.005,
1068
+ 0.02,
1069
+ 0.01,
1070
+ 0.005,
1071
+ 0.005,
1072
+ 0.005,
1073
+ 0.015,
1074
+ 0.015,
1075
+ 0.02,
1076
+ 0.0
1077
+ ],
1078
+ "elapsed_seconds": 191.17997980117798
1079
+ },
1080
+ {
1081
+ "group": "Old TRL",
1082
+ "experiment_id": "sb3_ppo_math_s1024",
1083
+ "experiment_name": "SB3 PPO Math seed=1024",
1084
+ "model": "SB3/PPO (policy network)",
1085
+ "model_short": "sb3-ppo",
1086
+ "method": "PPO",
1087
+ "task": "math",
1088
+ "platform": "modal",
1089
+ "seed": 1024,
1090
+ "steps_completed": 100352,
1091
+ "peak_reward": 0.025,
1092
+ "last10_avg": 0.01,
1093
+ "first5_avg": 0.003,
1094
+ "status": "completed",
1095
+ "error_message": null,
1096
+ "finding": "SB3/PPO: final accuracy=0.010 \u2014 policy gradient on raw math fails without LLM",
1097
+ "wandb_run_url": null,
1098
+ "hf_checkpoint_url": null,
1099
+ "reward_trace": [
1100
+ 0.0,
1101
+ 0.01,
1102
+ 0.005,
1103
+ 0.0,
1104
+ 0.0,
1105
+ 0.005,
1106
+ 0.0,
1107
+ 0.0,
1108
+ 0.0,
1109
+ 0.0,
1110
+ 0.005,
1111
+ 0.0,
1112
+ 0.01,
1113
+ 0.005,
1114
+ 0.005,
1115
+ 0.01,
1116
+ 0.015,
1117
+ 0.0,
1118
+ 0.005,
1119
+ 0.005,
1120
+ 0.015,
1121
+ 0.025,
1122
+ 0.005,
1123
+ 0.005,
1124
+ 0.005,
1125
+ 0.005,
1126
+ 0.005,
1127
+ 0.005,
1128
+ 0.0,
1129
+ 0.01,
1130
+ 0.0,
1131
+ 0.005,
1132
+ 0.005,
1133
+ 0.0,
1134
+ 0.005,
1135
+ 0.01,
1136
+ 0.005,
1137
+ 0.015,
1138
+ 0.0,
1139
+ 0.005,
1140
+ 0.0,
1141
+ 0.025,
1142
+ 0.02,
1143
+ 0.005,
1144
+ 0.01,
1145
+ 0.01,
1146
+ 0.01,
1147
+ 0.01,
1148
+ 0.005
1149
+ ],
1150
+ "elapsed_seconds": 187.7582619190216
1151
+ },
1152
+ {
1153
+ "group": "Old TRL",
1154
+ "experiment_id": "cleanrl_ppo_math_s42",
1155
+ "experiment_name": "CleanRL PPO Math seed=42",
1156
+ "model": "CleanRL/PPO (policy network)",
1157
+ "model_short": "cleanrl-ppo",
1158
+ "method": "PPO",
1159
+ "task": "math",
1160
+ "platform": "modal",
1161
+ "seed": 42,
1162
+ "steps_completed": 97280,
1163
+ "peak_reward": 0.02,
1164
+ "last10_avg": 0.011,
1165
+ "first5_avg": 0.008,
1166
+ "status": "completed",
1167
+ "error_message": null,
1168
+ "finding": "CleanRL/PPO: final accuracy=0.010 \u2014 near-zero, confirms LLM mandatory",
1169
+ "wandb_run_url": null,
1170
+ "hf_checkpoint_url": null,
1171
+ "reward_trace": [
1172
+ 0.0,
1173
+ 0.015,
1174
+ 0.015,
1175
+ 0.01,
1176
+ 0.0,
1177
+ 0.005,
1178
+ 0.015,
1179
+ 0.01,
1180
+ 0.015,
1181
+ 0.01,
1182
+ 0.02,
1183
+ 0.005,
1184
+ 0.01,
1185
+ 0.005,
1186
+ 0.005,
1187
+ 0.01,
1188
+ 0.02,
1189
+ 0.01,
1190
+ 0.015
1191
+ ],
1192
+ "elapsed_seconds": 25.303205966949463
1193
+ },
1194
+ {
1195
+ "group": "Old TRL",
1196
+ "experiment_id": "cleanrl_ppo_math_s123",
1197
+ "experiment_name": "CleanRL PPO Math seed=123",
1198
+ "model": "CleanRL/PPO (policy network)",
1199
+ "model_short": "cleanrl-ppo",
1200
+ "method": "PPO",
1201
+ "task": "math",
1202
+ "platform": "modal",
1203
+ "seed": 123,
1204
+ "steps_completed": 97280,
1205
+ "peak_reward": 0.005,
1206
+ "last10_avg": 0.002,
1207
+ "first5_avg": 0.0,
1208
+ "status": "completed",
1209
+ "error_message": null,
1210
+ "finding": "CleanRL/PPO: final accuracy=0.008 \u2014 near-zero, confirms LLM mandatory",
1211
+ "wandb_run_url": null,
1212
+ "hf_checkpoint_url": null,
1213
+ "reward_trace": [
1214
+ 0.0,
1215
+ 0.0,
1216
+ 0.0,
1217
+ 0.0,
1218
+ 0.0,
1219
+ 0.0,
1220
+ 0.0,
1221
+ 0.005,
1222
+ 0.005,
1223
+ 0.0,
1224
+ 0.0,
1225
+ 0.005,
1226
+ 0.0,
1227
+ 0.0,
1228
+ 0.0,
1229
+ 0.005,
1230
+ 0.0,
1231
+ 0.005,
1232
+ 0.005
1233
+ ],
1234
+ "elapsed_seconds": 26.161928415298462
1235
+ },
1236
+ {
1237
+ "group": "Old TRL",
1238
+ "experiment_id": "cleanrl_ppo_math_s456",
1239
+ "experiment_name": "CleanRL PPO Math seed=456",
1240
+ "model": "CleanRL/PPO (policy network)",
1241
+ "model_short": "cleanrl-ppo",
1242
+ "method": "PPO",
1243
+ "task": "math",
1244
+ "platform": "modal",
1245
+ "seed": 456,
1246
+ "steps_completed": 97280,
1247
+ "peak_reward": 0.015,
1248
+ "last10_avg": 0.008,
1249
+ "first5_avg": 0.005,
1250
+ "status": "completed",
1251
+ "error_message": null,
1252
+ "finding": "CleanRL/PPO: final accuracy=0.014 \u2014 near-zero, confirms LLM mandatory",
1253
+ "wandb_run_url": null,
1254
+ "hf_checkpoint_url": null,
1255
+ "reward_trace": [
1256
+ 0.0,
1257
+ 0.005,
1258
+ 0.01,
1259
+ 0.005,
1260
+ 0.005,
1261
+ 0.0,
1262
+ 0.01,
1263
+ 0.005,
1264
+ 0.01,
1265
+ 0.005,
1266
+ 0.01,
1267
+ 0.005,
1268
+ 0.015,
1269
+ 0.005,
1270
+ 0.005,
1271
+ 0.005,
1272
+ 0.015,
1273
+ 0.005,
1274
+ 0.01
1275
+ ],
1276
+ "elapsed_seconds": 24.993607997894287
1277
+ },
1278
+ {
1279
+ "group": "Old TRL",
1280
+ "experiment_id": "cleanrl_ppo_math_s789",
1281
+ "experiment_name": "CleanRL PPO Math seed=789",
1282
+ "model": "CleanRL/PPO (policy network)",
1283
+ "model_short": "cleanrl-ppo",
1284
+ "method": "PPO",
1285
+ "task": "math",
1286
+ "platform": "modal",
1287
+ "seed": 789,
1288
+ "steps_completed": 97280,
1289
+ "peak_reward": 0.025,
1290
+ "last10_avg": 0.0095,
1291
+ "first5_avg": 0.012,
1292
+ "status": "completed",
1293
+ "error_message": null,
1294
+ "finding": "CleanRL/PPO: final accuracy=0.009 \u2014 near-zero, confirms LLM mandatory",
1295
+ "wandb_run_url": null,
1296
+ "hf_checkpoint_url": null,
1297
+ "reward_trace": [
1298
+ 0.01,
1299
+ 0.0,
1300
+ 0.01,
1301
+ 0.015,
1302
+ 0.025,
1303
+ 0.015,
1304
+ 0.015,
1305
+ 0.005,
1306
+ 0.0,
1307
+ 0.015,
1308
+ 0.01,
1309
+ 0.005,
1310
+ 0.015,
1311
+ 0.01,
1312
+ 0.01,
1313
+ 0.01,
1314
+ 0.005,
1315
+ 0.005,
1316
+ 0.01
1317
+ ],
1318
+ "elapsed_seconds": 25.0768883228302
1319
+ },
1320
+ {
1321
+ "group": "Old TRL",
1322
+ "experiment_id": "cleanrl_ppo_math_s1024",
1323
+ "experiment_name": "CleanRL PPO Math seed=1024",
1324
+ "model": "CleanRL/PPO (policy network)",
1325
+ "model_short": "cleanrl-ppo",
1326
+ "method": "PPO",
1327
+ "task": "math",
1328
+ "platform": "modal",
1329
+ "seed": 1024,
1330
+ "steps_completed": 97280,
1331
+ "peak_reward": 0.01,
1332
+ "last10_avg": 0.0025,
1333
+ "first5_avg": 0.004,
1334
+ "status": "completed",
1335
+ "error_message": null,
1336
+ "finding": "CleanRL/PPO: final accuracy=0.004 \u2014 near-zero, confirms LLM mandatory",
1337
+ "wandb_run_url": null,
1338
+ "hf_checkpoint_url": null,
1339
+ "reward_trace": [
1340
+ 0.005,
1341
+ 0.0,
1342
+ 0.005,
1343
+ 0.0,
1344
+ 0.01,
1345
+ 0.005,
1346
+ 0.005,
1347
+ 0.005,
1348
+ 0.0,
1349
+ 0.0,
1350
+ 0.005,
1351
+ 0.0,
1352
+ 0.005,
1353
+ 0.005,
1354
+ 0.0,
1355
+ 0.0,
1356
+ 0.01,
1357
+ 0.0,
1358
+ 0.0
1359
+ ],
1360
+ "elapsed_seconds": 32.029313802719116
1361
+ },
1362
+ {
1363
+ "group": "Old TRL",
1364
+ "experiment_id": "tianshou_ppo_math_s42",
1365
+ "experiment_name": "Tianshou PPO Math seed=42",
1366
+ "model": "Tianshou/PPO (policy network)",
1367
+ "model_short": "tianshou-ppo",
1368
+ "method": "PPO",
1369
+ "task": "math",
1370
+ "platform": "modal",
1371
+ "seed": 42,
1372
+ "steps_completed": 100000,
1373
+ "peak_reward": 0.025,
1374
+ "last10_avg": 0.0135,
1375
+ "first5_avg": 0.001,
1376
+ "status": "completed",
1377
+ "error_message": null,
1378
+ "finding": "Tianshou/PPO: final accuracy=0.011 \u2014 near-zero, <3s runtime indicates trivial baseline",
1379
+ "wandb_run_url": null,
1380
+ "hf_checkpoint_url": null,
1381
+ "reward_trace": [
1382
+ 0.0,
1383
+ 0.005,
1384
+ 0.0,
1385
+ 0.0,
1386
+ 0.0,
1387
+ 0.0,
1388
+ 0.005,
1389
+ 0.005,
1390
+ 0.005,
1391
+ 0.005,
1392
+ 0.01,
1393
+ 0.01,
1394
+ 0.0,
1395
+ 0.015,
1396
+ 0.005,
1397
+ 0.025,
1398
+ 0.02,
1399
+ 0.025,
1400
+ 0.01,
1401
+ 0.015
1402
+ ],
1403
+ "elapsed_seconds": 2.7088005542755127
1404
+ },
1405
+ {
1406
+ "group": "Old TRL",
1407
+ "experiment_id": "tianshou_ppo_math_s123",
1408
+ "experiment_name": "Tianshou PPO Math seed=123",
1409
+ "model": "Tianshou/PPO (policy network)",
1410
+ "model_short": "tianshou-ppo",
1411
+ "method": "PPO",
1412
+ "task": "math",
1413
+ "platform": "modal",
1414
+ "seed": 123,
1415
+ "steps_completed": 100000,
1416
+ "peak_reward": 0.02,
1417
+ "last10_avg": 0.0055,
1418
+ "first5_avg": 0.002,
1419
+ "status": "completed",
1420
+ "error_message": null,
1421
+ "finding": "Tianshou/PPO: final accuracy=0.002 \u2014 near-zero, <3s runtime indicates trivial baseline",
1422
+ "wandb_run_url": null,
1423
+ "hf_checkpoint_url": null,
1424
+ "reward_trace": [
1425
+ 0.0,
1426
+ 0.005,
1427
+ 0.0,
1428
+ 0.0,
1429
+ 0.005,
1430
+ 0.0,
1431
+ 0.0,
1432
+ 0.01,
1433
+ 0.01,
1434
+ 0.0,
1435
+ 0.0,
1436
+ 0.02,
1437
+ 0.005,
1438
+ 0.005,
1439
+ 0.0,
1440
+ 0.005,
1441
+ 0.0,
1442
+ 0.0,
1443
+ 0.01,
1444
+ 0.01
1445
+ ],
1446
+ "elapsed_seconds": 2.612163782119751
1447
+ },
1448
+ {
1449
+ "group": "Old TRL",
1450
+ "experiment_id": "tianshou_ppo_math_s456",
1451
+ "experiment_name": "Tianshou PPO Math seed=456",
1452
+ "model": "Tianshou/PPO (policy network)",
1453
+ "model_short": "tianshou-ppo",
1454
+ "method": "PPO",
1455
+ "task": "math",
1456
+ "platform": "modal",
1457
+ "seed": 456,
1458
+ "steps_completed": 100000,
1459
+ "peak_reward": 0.02,
1460
+ "last10_avg": 0.01,
1461
+ "first5_avg": 0.007,
1462
+ "status": "completed",
1463
+ "error_message": null,
1464
+ "finding": "Tianshou/PPO: final accuracy=0.005 \u2014 near-zero, <3s runtime indicates trivial baseline",
1465
+ "wandb_run_url": null,
1466
+ "hf_checkpoint_url": null,
1467
+ "reward_trace": [
1468
+ 0.01,
1469
+ 0.01,
1470
+ 0.005,
1471
+ 0.005,
1472
+ 0.005,
1473
+ 0.005,
1474
+ 0.015,
1475
+ 0.01,
1476
+ 0.005,
1477
+ 0.005,
1478
+ 0.01,
1479
+ 0.0,
1480
+ 0.02,
1481
+ 0.015,
1482
+ 0.01,
1483
+ 0.01,
1484
+ 0.015,
1485
+ 0.005,
1486
+ 0.01,
1487
+ 0.005
1488
+ ],
1489
+ "elapsed_seconds": 3.0149343013763428
1490
+ },
1491
+ {
1492
+ "group": "Old TRL",
1493
+ "experiment_id": "tianshou_ppo_math_s789",
1494
+ "experiment_name": "Tianshou PPO Math seed=789",
1495
+ "model": "Tianshou/PPO (policy network)",
1496
+ "model_short": "tianshou-ppo",
1497
+ "method": "PPO",
1498
+ "task": "math",
1499
+ "platform": "modal",
1500
+ "seed": 789,
1501
+ "steps_completed": 100000,
1502
+ "peak_reward": 0.02,
1503
+ "last10_avg": 0.006,
1504
+ "first5_avg": 0.005,
1505
+ "status": "completed",
1506
+ "error_message": null,
1507
+ "finding": "Tianshou/PPO: final accuracy=0.002 \u2014 near-zero, <3s runtime indicates trivial baseline",
1508
+ "wandb_run_url": null,
1509
+ "hf_checkpoint_url": null,
1510
+ "reward_trace": [
1511
+ 0.0,
1512
+ 0.005,
1513
+ 0.015,
1514
+ 0.005,
1515
+ 0.0,
1516
+ 0.0,
1517
+ 0.02,
1518
+ 0.0,
1519
+ 0.01,
1520
+ 0.005,
1521
+ 0.01,
1522
+ 0.005,
1523
+ 0.005,
1524
+ 0.01,
1525
+ 0.0,
1526
+ 0.0,
1527
+ 0.01,
1528
+ 0.005,
1529
+ 0.015,
1530
+ 0.0
1531
+ ],
1532
+ "elapsed_seconds": 3.2362585067749023
1533
+ },
1534
+ {
1535
+ "group": "Old TRL",
1536
+ "experiment_id": "tianshou_ppo_math_s1024",
1537
+ "experiment_name": "Tianshou PPO Math seed=1024",
1538
+ "model": "Tianshou/PPO (policy network)",
1539
+ "model_short": "tianshou-ppo",
1540
+ "method": "PPO",
1541
+ "task": "math",
1542
+ "platform": "modal",
1543
+ "seed": 1024,
1544
+ "steps_completed": 100000,
1545
+ "peak_reward": 0.02,
1546
+ "last10_avg": 0.0075,
1547
+ "first5_avg": 0.007,
1548
+ "status": "completed",
1549
+ "error_message": null,
1550
+ "finding": "Tianshou/PPO: final accuracy=0.009 \u2014 near-zero, <3s runtime indicates trivial baseline",
1551
+ "wandb_run_url": null,
1552
+ "hf_checkpoint_url": null,
1553
+ "reward_trace": [
1554
+ 0.015,
1555
+ 0.0,
1556
+ 0.005,
1557
+ 0.005,
1558
+ 0.01,
1559
+ 0.005,
1560
+ 0.01,
1561
+ 0.015,
1562
+ 0.005,
1563
+ 0.005,
1564
+ 0.02,
1565
+ 0.0,
1566
+ 0.0,
1567
+ 0.01,
1568
+ 0.01,
1569
+ 0.01,
1570
+ 0.005,
1571
+ 0.015,
1572
+ 0.0,
1573
+ 0.005
1574
+ ],
1575
+ "elapsed_seconds": 3.208648681640625
1576
+ },
1577
+ {
1578
+ "group": "Team Member",
1579
+ "experiment_id": "sandhya_grpo_tool_calling_3b",
1580
+ "experiment_name": "Sandhya \u2013 GRPO Tool Calling (3B)",
1581
+ "model": "3B (unspecified base)",
1582
+ "model_short": "3b-tool",
1583
+ "method": "GRPO vs SFT",
1584
+ "task": "tool_calling",
1585
+ "platform": "external",
1586
+ "seed": null,
1587
+ "steps_completed": null,
1588
+ "peak_reward": 0.91,
1589
+ "last10_avg": 0.91,
1590
+ "first5_avg": null,
1591
+ "status": "completed",
1592
+ "error_message": null,
1593
+ "finding": "GRPO 0.91 vs SFT baseline 0.72 on tool-calling \u2014 +0.19 absolute improvement with RL",
1594
+ "wandb_run_url": null,
1595
+ "hf_checkpoint_url": null,
1596
+ "reward_trace": [],
1597
+ "sft_baseline": 0.72,
1598
+ "delta": "+0.19"
1599
+ },
1600
+ {
1601
+ "group": "Team Member",
1602
+ "experiment_id": "madhu_grpo_humaneval_qwen3-8b",
1603
+ "experiment_name": "Madhu \u2013 HumanEval (Qwen3-8B)",
1604
+ "model": "Qwen/Qwen3-8B",
1605
+ "model_short": "qwen3-8b",
1606
+ "method": "GRPO",
1607
+ "task": "code_generation",
1608
+ "platform": "external",
1609
+ "seed": null,
1610
+ "steps_completed": null,
1611
+ "peak_reward": 0.86,
1612
+ "last10_avg": 0.86,
1613
+ "first5_avg": null,
1614
+ "status": "completed",
1615
+ "error_message": null,
1616
+ "finding": "HumanEval 86% (141/164 problems) \u2014 best code-gen result in the project",
1617
+ "wandb_run_url": null,
1618
+ "hf_checkpoint_url": "https://github.com/madhukumara1993/qwen3-grpo",
1619
+ "reward_trace": [],
1620
+ "humaneval_score": "86% (141/164)"
1621
+ },
1622
+ {
1623
+ "group": "Team Member",
1624
+ "experiment_id": "mohammad_rafi_grpo_math_qwen3-4b",
1625
+ "experiment_name": "Mohammad Rafi \u2013 Math Reasoning (Qwen3-4B)",
1626
+ "model": "Qwen/Qwen3-4B",
1627
+ "model_short": "qwen3-4b",
1628
+ "method": "GRPO",
1629
+ "task": "math_reasoning",
1630
+ "platform": "external",
1631
+ "seed": null,
1632
+ "steps_completed": null,
1633
+ "peak_reward": 0.678,
1634
+ "last10_avg": 0.678,
1635
+ "first5_avg": null,
1636
+ "status": "completed",
1637
+ "error_message": null,
1638
+ "finding": "+0.6pp accuracy gain (67.2% \u2192 67.8%) on math reasoning with Qwen3-4B GRPO",
1639
+ "wandb_run_url": null,
1640
+ "hf_checkpoint_url": null,
1641
+ "reward_trace": [],
1642
+ "baseline_accuracy": 0.672,
1643
+ "final_accuracy": 0.678,
1644
+ "delta": "+0.6pp"
1645
+ },
1646
+ {
1647
+ "group": "Team Member",
1648
+ "experiment_id": "arumugam_dpo_keyword_0.5b",
1649
+ "experiment_name": "Arumugam \u2013 DPO Keyword Eval (0.5B)",
1650
+ "model": "0.5B (unspecified base)",
1651
+ "model_short": "0.5b-dpo",
1652
+ "method": "DPO",
1653
+ "task": "keyword_generation",
1654
+ "platform": "external",
1655
+ "seed": null,
1656
+ "steps_completed": null,
1657
+ "peak_reward": null,
1658
+ "last10_avg": null,
1659
+ "first5_avg": null,
1660
+ "status": "completed",
1661
+ "error_message": null,
1662
+ "finding": "+25% keyword metric with DPO on 0.5B model (8 training examples \u2014 very low-data regime)",
1663
+ "wandb_run_url": null,
1664
+ "hf_checkpoint_url": null,
1665
+ "reward_trace": [],
1666
+ "keyword_delta": "+25%",
1667
+ "training_examples": 8,
1668
+ "note": "8 training examples, keyword eval \u2014 low-data regime, limited generalizability"
1669
+ },
1670
+ {
1671
+ "experiment": "cross_tool_llama-8b-inst",
1672
+ "model": "meta-llama/Llama-3.1-8B-Instruct",
1673
+ "model_short": "llama-8b-inst",
1674
+ "task": "tool_use",
1675
+ "seed": 42,
1676
+ "rank": 32,
1677
+ "lr": 3e-05,
1678
+ "group": 8,
1679
+ "steps": 30,
1680
+ "run_id": "ca2e3a24-7401-5770-af34-a0d27177aeaa:train:0",
1681
+ "checkpoint": "tinker://ca2e3a24-7401-5770-af34-a0d27177aeaa:train:0/sampler_weights/final",
1682
+ "first5_avg": 0.0,
1683
+ "last10_avg": 0.0,
1684
+ "peak": 0.0,
1685
+ "zero_loss_pct": 100.0,
1686
+ "zero_reward_pct": 100.0,
1687
+ "reward_trace": [
1688
+ 0.0,
1689
+ 0.0,
1690
+ 0.0,
1691
+ 0.0,
1692
+ 0.0,
1693
+ 0.0,
1694
+ 0.0,
1695
+ 0.0,
1696
+ 0.0,
1697
+ 0.0,
1698
+ 0.0,
1699
+ 0.0,
1700
+ 0.0,
1701
+ 0.0,
1702
+ 0.0,
1703
+ 0.0,
1704
+ 0.0,
1705
+ 0.0,
1706
+ 0.0,
1707
+ 0.0,
1708
+ 0.0,
1709
+ 0.0,
1710
+ 0.0,
1711
+ 0.0,
1712
+ 0.0,
1713
+ 0.0,
1714
+ 0.0,
1715
+ 0.0,
1716
+ 0.0,
1717
+ 0.0
1718
+ ],
1719
+ "step_log": [
1720
+ {
1721
+ "step": 1,
1722
+ "loss": 0.0,
1723
+ "reward": 0.0,
1724
+ "zvf": 1.0,
1725
+ "gu": 0.0
1726
+ },
1727
+ {
1728
+ "step": 2,
1729
+ "loss": 0.0,
1730
+ "reward": 0.0,
1731
+ "zvf": 1.0,
1732
+ "gu": 0.0
1733
+ },
1734
+ {
1735
+ "step": 3,
1736
+ "loss": 0.0,
1737
+ "reward": 0.0,
1738
+ "zvf": 1.0,
1739
+ "gu": 0.0
1740
+ },
1741
+ {
1742
+ "step": 4,
1743
+ "loss": 0.0,
1744
+ "reward": 0.0,
1745
+ "zvf": 1.0,
1746
+ "gu": 0.0
1747
+ },
1748
+ {
1749
+ "step": 5,
1750
+ "loss": 0.0,
1751
+ "reward": 0.0,
1752
+ "zvf": 1.0,
1753
+ "gu": 0.0
1754
+ },
1755
+ {
1756
+ "step": 6,
1757
+ "loss": 0.0,
1758
+ "reward": 0.0,
1759
+ "zvf": 1.0,
1760
+ "gu": 0.0
1761
+ },
1762
+ {
1763
+ "step": 7,
1764
+ "loss": 0.0,
1765
+ "reward": 0.0,
1766
+ "zvf": 1.0,
1767
+ "gu": 0.0
1768
+ },
1769
+ {
1770
+ "step": 8,
1771
+ "loss": 0.0,
1772
+ "reward": 0.0,
1773
+ "zvf": 1.0,
1774
+ "gu": 0.0
1775
+ },
1776
+ {
1777
+ "step": 9,
1778
+ "loss": 0.0,
1779
+ "reward": 0.0,
1780
+ "zvf": 1.0,
1781
+ "gu": 0.0
1782
+ },
1783
+ {
1784
+ "step": 10,
1785
+ "loss": 0.0,
1786
+ "reward": 0.0,
1787
+ "zvf": 1.0,
1788
+ "gu": 0.0
1789
+ },
1790
+ {
1791
+ "step": 11,
1792
+ "loss": 0.0,
1793
+ "reward": 0.0,
1794
+ "zvf": 1.0,
1795
+ "gu": 0.0
1796
+ },
1797
+ {
1798
+ "step": 12,
1799
+ "loss": 0.0,
1800
+ "reward": 0.0,
1801
+ "zvf": 1.0,
1802
+ "gu": 0.0
1803
+ },
1804
+ {
1805
+ "step": 13,
1806
+ "loss": 0.0,
1807
+ "reward": 0.0,
1808
+ "zvf": 1.0,
1809
+ "gu": 0.0
1810
+ },
1811
+ {
1812
+ "step": 14,
1813
+ "loss": 0.0,
1814
+ "reward": 0.0,
1815
+ "zvf": 1.0,
1816
+ "gu": 0.0
1817
+ },
1818
+ {
1819
+ "step": 15,
1820
+ "loss": 0.0,
1821
+ "reward": 0.0,
1822
+ "zvf": 1.0,
1823
+ "gu": 0.0
1824
+ },
1825
+ {
1826
+ "step": 16,
1827
+ "loss": 0.0,
1828
+ "reward": 0.0,
1829
+ "zvf": 1.0,
1830
+ "gu": 0.0
1831
+ },
1832
+ {
1833
+ "step": 17,
1834
+ "loss": 0.0,
1835
+ "reward": 0.0,
1836
+ "zvf": 1.0,
1837
+ "gu": 0.0
1838
+ },
1839
+ {
1840
+ "step": 18,
1841
+ "loss": 0.0,
1842
+ "reward": 0.0,
1843
+ "zvf": 1.0,
1844
+ "gu": 0.0
1845
+ },
1846
+ {
1847
+ "step": 19,
1848
+ "loss": 0.0,
1849
+ "reward": 0.0,
1850
+ "zvf": 1.0,
1851
+ "gu": 0.0
1852
+ },
1853
+ {
1854
+ "step": 20,
1855
+ "loss": 0.0,
1856
+ "reward": 0.0,
1857
+ "zvf": 1.0,
1858
+ "gu": 0.0
1859
+ },
1860
+ {
1861
+ "step": 21,
1862
+ "loss": 0.0,
1863
+ "reward": 0.0,
1864
+ "zvf": 1.0,
1865
+ "gu": 0.0
1866
+ },
1867
+ {
1868
+ "step": 22,
1869
+ "loss": 0.0,
1870
+ "reward": 0.0,
1871
+ "zvf": 1.0,
1872
+ "gu": 0.0
1873
+ },
1874
+ {
1875
+ "step": 23,
1876
+ "loss": 0.0,
1877
+ "reward": 0.0,
1878
+ "zvf": 1.0,
1879
+ "gu": 0.0
1880
+ },
1881
+ {
1882
+ "step": 24,
1883
+ "loss": 0.0,
1884
+ "reward": 0.0,
1885
+ "zvf": 1.0,
1886
+ "gu": 0.0
1887
+ },
1888
+ {
1889
+ "step": 25,
1890
+ "loss": 0.0,
1891
+ "reward": 0.0,
1892
+ "zvf": 1.0,
1893
+ "gu": 0.0
1894
+ },
1895
+ {
1896
+ "step": 26,
1897
+ "loss": 0.0,
1898
+ "reward": 0.0,
1899
+ "zvf": 1.0,
1900
+ "gu": 0.0
1901
+ },
1902
+ {
1903
+ "step": 27,
1904
+ "loss": 0.0,
1905
+ "reward": 0.0,
1906
+ "zvf": 1.0,
1907
+ "gu": 0.0
1908
+ },
1909
+ {
1910
+ "step": 28,
1911
+ "loss": 0.0,
1912
+ "reward": 0.0,
1913
+ "zvf": 1.0,
1914
+ "gu": 0.0
1915
+ },
1916
+ {
1917
+ "step": 29,
1918
+ "loss": 0.0,
1919
+ "reward": 0.0,
1920
+ "zvf": 1.0,
1921
+ "gu": 0.0
1922
+ },
1923
+ {
1924
+ "step": 30,
1925
+ "loss": 0.0,
1926
+ "reward": 0.0,
1927
+ "zvf": 1.0,
1928
+ "gu": 0.0
1929
+ }
1930
+ ],
1931
+ "timestamp": "2026-04-18T16:01:52.013977",
1932
+ "platform": "tinker",
1933
+ "peak_accuracy": 0.0,
1934
+ "last10_accuracy": 0.0
1935
+ },
1936
+ {
1937
+ "experiment": "cross_tool_qwen3-32b",
1938
+ "model": "Qwen/Qwen3-32B",
1939
+ "model_short": "qwen3-32b",
1940
+ "task": "tool_use",
1941
+ "seed": 42,
1942
+ "rank": 32,
1943
+ "lr": 3e-05,
1944
+ "group": 8,
1945
+ "steps": 30,
1946
+ "run_id": "73ff186a-d8e3-50c3-afb8-2b863cd09579:train:0",
1947
+ "checkpoint": "tinker://73ff186a-d8e3-50c3-afb8-2b863cd09579:train:0/sampler_weights/final",
1948
+ "first5_avg": 0.0,
1949
+ "last10_avg": 0.0,
1950
+ "peak": 0.0,
1951
+ "zero_loss_pct": 100.0,
1952
+ "zero_reward_pct": 100.0,
1953
+ "reward_trace": [
1954
+ 0.0,
1955
+ 0.0,
1956
+ 0.0,
1957
+ 0.0,
1958
+ 0.0,
1959
+ 0.0,
1960
+ 0.0,
1961
+ 0.0,
1962
+ 0.0,
1963
+ 0.0,
1964
+ 0.0,
1965
+ 0.0,
1966
+ 0.0,
1967
+ 0.0,
1968
+ 0.0,
1969
+ 0.0,
1970
+ 0.0,
1971
+ 0.0,
1972
+ 0.0,
1973
+ 0.0,
1974
+ 0.0,
1975
+ 0.0,
1976
+ 0.0,
1977
+ 0.0,
1978
+ 0.0,
1979
+ 0.0,
1980
+ 0.0,
1981
+ 0.0,
1982
+ 0.0,
1983
+ 0.0
1984
+ ],
1985
+ "step_log": [
1986
+ {
1987
+ "step": 1,
1988
+ "loss": 0.0,
1989
+ "reward": 0.0,
1990
+ "zvf": 1.0,
1991
+ "gu": 0.0
1992
+ },
1993
+ {
1994
+ "step": 2,
1995
+ "loss": 0.0,
1996
+ "reward": 0.0,
1997
+ "zvf": 1.0,
1998
+ "gu": 0.0
1999
+ },
2000
+ {
2001
+ "step": 3,
2002
+ "loss": 0.0,
2003
+ "reward": 0.0,
2004
+ "zvf": 1.0,
2005
+ "gu": 0.0
2006
+ },
2007
+ {
2008
+ "step": 4,
2009
+ "loss": 0.0,
2010
+ "reward": 0.0,
2011
+ "zvf": 1.0,
2012
+ "gu": 0.0
2013
+ },
2014
+ {
2015
+ "step": 5,
2016
+ "loss": 0.0,
2017
+ "reward": 0.0,
2018
+ "zvf": 1.0,
2019
+ "gu": 0.0
2020
+ },
2021
+ {
2022
+ "step": 6,
2023
+ "loss": 0.0,
2024
+ "reward": 0.0,
2025
+ "zvf": 1.0,
2026
+ "gu": 0.0
2027
+ },
2028
+ {
2029
+ "step": 7,
2030
+ "loss": 0.0,
2031
+ "reward": 0.0,
2032
+ "zvf": 1.0,
2033
+ "gu": 0.0
2034
+ },
2035
+ {
2036
+ "step": 8,
2037
+ "loss": 0.0,
2038
+ "reward": 0.0,
2039
+ "zvf": 1.0,
2040
+ "gu": 0.0
2041
+ },
2042
+ {
2043
+ "step": 9,
2044
+ "loss": 0.0,
2045
+ "reward": 0.0,
2046
+ "zvf": 1.0,
2047
+ "gu": 0.0
2048
+ },
2049
+ {
2050
+ "step": 10,
2051
+ "loss": 0.0,
2052
+ "reward": 0.0,
2053
+ "zvf": 1.0,
2054
+ "gu": 0.0
2055
+ },
2056
+ {
2057
+ "step": 11,
2058
+ "loss": 0.0,
2059
+ "reward": 0.0,
2060
+ "zvf": 1.0,
2061
+ "gu": 0.0
2062
+ },
2063
+ {
2064
+ "step": 12,
2065
+ "loss": 0.0,
2066
+ "reward": 0.0,
2067
+ "zvf": 1.0,
2068
+ "gu": 0.0
2069
+ },
2070
+ {
2071
+ "step": 13,
2072
+ "loss": 0.0,
2073
+ "reward": 0.0,
2074
+ "zvf": 1.0,
2075
+ "gu": 0.0
2076
+ },
2077
+ {
2078
+ "step": 14,
2079
+ "loss": 0.0,
2080
+ "reward": 0.0,
2081
+ "zvf": 1.0,
2082
+ "gu": 0.0
2083
+ },
2084
+ {
2085
+ "step": 15,
2086
+ "loss": 0.0,
2087
+ "reward": 0.0,
2088
+ "zvf": 1.0,
2089
+ "gu": 0.0
2090
+ },
2091
+ {
2092
+ "step": 16,
2093
+ "loss": 0.0,
2094
+ "reward": 0.0,
2095
+ "zvf": 1.0,
2096
+ "gu": 0.0
2097
+ },
2098
+ {
2099
+ "step": 17,
2100
+ "loss": 0.0,
2101
+ "reward": 0.0,
2102
+ "zvf": 1.0,
2103
+ "gu": 0.0
2104
+ },
2105
+ {
2106
+ "step": 18,
2107
+ "loss": 0.0,
2108
+ "reward": 0.0,
2109
+ "zvf": 1.0,
2110
+ "gu": 0.0
2111
+ },
2112
+ {
2113
+ "step": 19,
2114
+ "loss": 0.0,
2115
+ "reward": 0.0,
2116
+ "zvf": 1.0,
2117
+ "gu": 0.0
2118
+ },
2119
+ {
2120
+ "step": 20,
2121
+ "loss": 0.0,
2122
+ "reward": 0.0,
2123
+ "zvf": 1.0,
2124
+ "gu": 0.0
2125
+ },
2126
+ {
2127
+ "step": 21,
2128
+ "loss": 0.0,
2129
+ "reward": 0.0,
2130
+ "zvf": 1.0,
2131
+ "gu": 0.0
2132
+ },
2133
+ {
2134
+ "step": 22,
2135
+ "loss": 0.0,
2136
+ "reward": 0.0,
2137
+ "zvf": 1.0,
2138
+ "gu": 0.0
2139
+ },
2140
+ {
2141
+ "step": 23,
2142
+ "loss": 0.0,
2143
+ "reward": 0.0,
2144
+ "zvf": 1.0,
2145
+ "gu": 0.0
2146
+ },
2147
+ {
2148
+ "step": 24,
2149
+ "loss": 0.0,
2150
+ "reward": 0.0,
2151
+ "zvf": 1.0,
2152
+ "gu": 0.0
2153
+ },
2154
+ {
2155
+ "step": 25,
2156
+ "loss": 0.0,
2157
+ "reward": 0.0,
2158
+ "zvf": 1.0,
2159
+ "gu": 0.0
2160
+ },
2161
+ {
2162
+ "step": 26,
2163
+ "loss": 0.0,
2164
+ "reward": 0.0,
2165
+ "zvf": 1.0,
2166
+ "gu": 0.0
2167
+ },
2168
+ {
2169
+ "step": 27,
2170
+ "loss": 0.0,
2171
+ "reward": 0.0,
2172
+ "zvf": 1.0,
2173
+ "gu": 0.0
2174
+ },
2175
+ {
2176
+ "step": 28,
2177
+ "loss": 0.0,
2178
+ "reward": 0.0,
2179
+ "zvf": 1.0,
2180
+ "gu": 0.0
2181
+ },
2182
+ {
2183
+ "step": 29,
2184
+ "loss": 0.0,
2185
+ "reward": 0.0,
2186
+ "zvf": 1.0,
2187
+ "gu": 0.0
2188
+ },
2189
+ {
2190
+ "step": 30,
2191
+ "loss": 0.0,
2192
+ "reward": 0.0,
2193
+ "zvf": 1.0,
2194
+ "gu": 0.0
2195
+ }
2196
+ ],
2197
+ "timestamp": "2026-04-19T03:41:34.408077",
2198
+ "platform": "tinker",
2199
+ "peak_accuracy": 0.0,
2200
+ "last10_accuracy": 0.0
2201
+ },
2202
+ {
2203
+ "experiment": "frontier_gsm8k_deepseek-v3.1",
2204
+ "model": "deepseek-ai/DeepSeek-V3.1",
2205
+ "model_short": "deepseek-v3.1",
2206
+ "task": "gsm8k",
2207
+ "seed": 42,
2208
+ "rank": 16,
2209
+ "lr": 1e-05,
2210
+ "group": 4,
2211
+ "steps": 20,
2212
+ "run_id": "56b99b24-03bd-5d00-ae23-831933ef53b2:train:0",
2213
+ "checkpoint": "tinker://56b99b24-03bd-5d00-ae23-831933ef53b2:train:0/sampler_weights/final",
2214
+ "first5_avg": 0.85,
2215
+ "last10_avg": 0.85,
2216
+ "peak": 1.0,
2217
+ "zero_loss_pct": 30.0,
2218
+ "zero_reward_pct": 0.0,
2219
+ "reward_trace": [
2220
+ 0.875,
2221
+ 0.875,
2222
+ 1.0,
2223
+ 0.75,
2224
+ 0.75,
2225
+ 0.75,
2226
+ 0.625,
2227
+ 0.875,
2228
+ 0.875,
2229
+ 1.0,
2230
+ 0.75,
2231
+ 1.0,
2232
+ 1.0,
2233
+ 0.875,
2234
+ 1.0,
2235
+ 0.5,
2236
+ 0.875,
2237
+ 0.75,
2238
+ 0.75,
2239
+ 1.0
2240
+ ],
2241
+ "step_log": [
2242
+ {
2243
+ "step": 1,
2244
+ "loss": -9.515472412109375,
2245
+ "reward": 0.875,
2246
+ "zvf": 0.0,
2247
+ "gu": 1.0
2248
+ },
2249
+ {
2250
+ "step": 2,
2251
+ "loss": -9.95474624633789,
2252
+ "reward": 0.875,
2253
+ "zvf": 0.0,
2254
+ "gu": 1.0
2255
+ },
2256
+ {
2257
+ "step": 3,
2258
+ "loss": 0.0,
2259
+ "reward": 1.0,
2260
+ "zvf": 1.0,
2261
+ "gu": 0.0
2262
+ },
2263
+ {
2264
+ "step": 4,
2265
+ "loss": -8.243820190429688,
2266
+ "reward": 0.75,
2267
+ "zvf": 0.0,
2268
+ "gu": 1.0
2269
+ },
2270
+ {
2271
+ "step": 5,
2272
+ "loss": -4.6991777420043945,
2273
+ "reward": 0.75,
2274
+ "zvf": 0.0,
2275
+ "gu": 1.0
2276
+ },
2277
+ {
2278
+ "step": 6,
2279
+ "loss": -11.01620864868164,
2280
+ "reward": 0.75,
2281
+ "zvf": 0.0,
2282
+ "gu": 1.0
2283
+ },
2284
+ {
2285
+ "step": 7,
2286
+ "loss": 7.589530944824219,
2287
+ "reward": 0.625,
2288
+ "zvf": 0.0,
2289
+ "gu": 1.0
2290
+ },
2291
+ {
2292
+ "step": 8,
2293
+ "loss": -16.8707275390625,
2294
+ "reward": 0.875,
2295
+ "zvf": 0.0,
2296
+ "gu": 1.0
2297
+ },
2298
+ {
2299
+ "step": 9,
2300
+ "loss": -13.780792236328125,
2301
+ "reward": 0.875,
2302
+ "zvf": 0.0,
2303
+ "gu": 1.0
2304
+ },
2305
+ {
2306
+ "step": 10,
2307
+ "loss": 0.0,
2308
+ "reward": 1.0,
2309
+ "zvf": 1.0,
2310
+ "gu": 0.0
2311
+ },
2312
+ {
2313
+ "step": 11,
2314
+ "loss": -19.474306106567383,
2315
+ "reward": 0.75,
2316
+ "zvf": 0.0,
2317
+ "gu": 1.0
2318
+ },
2319
+ {
2320
+ "step": 12,
2321
+ "loss": 0.0,
2322
+ "reward": 1.0,
2323
+ "zvf": 1.0,
2324
+ "gu": 0.0
2325
+ },
2326
+ {
2327
+ "step": 13,
2328
+ "loss": 0.0,
2329
+ "reward": 1.0,
2330
+ "zvf": 1.0,
2331
+ "gu": 0.0
2332
+ },
2333
+ {
2334
+ "step": 14,
2335
+ "loss": -12.405517578125,
2336
+ "reward": 0.875,
2337
+ "zvf": 0.0,
2338
+ "gu": 1.0
2339
+ },
2340
+ {
2341
+ "step": 15,
2342
+ "loss": 0.0,
2343
+ "reward": 1.0,
2344
+ "zvf": 1.0,
2345
+ "gu": 0.0
2346
+ },
2347
+ {
2348
+ "step": 16,
2349
+ "loss": -14.82606315612793,
2350
+ "reward": 0.5,
2351
+ "zvf": 0.0,
2352
+ "gu": 1.0
2353
+ },
2354
+ {
2355
+ "step": 17,
2356
+ "loss": -12.409809112548828,
2357
+ "reward": 0.875,
2358
+ "zvf": 0.0,
2359
+ "gu": 1.0
2360
+ },
2361
+ {
2362
+ "step": 18,
2363
+ "loss": -13.506965637207031,
2364
+ "reward": 0.75,
2365
+ "zvf": 0.0,
2366
+ "gu": 1.0
2367
+ },
2368
+ {
2369
+ "step": 19,
2370
+ "loss": -14.026712417602539,
2371
+ "reward": 0.75,
2372
+ "zvf": 0.0,
2373
+ "gu": 1.0
2374
+ },
2375
+ {
2376
+ "step": 20,
2377
+ "loss": 0.0,
2378
+ "reward": 1.0,
2379
+ "zvf": 1.0,
2380
+ "gu": 0.0
2381
+ }
2382
+ ],
2383
+ "timestamp": "2026-04-18T16:05:46.209173",
2384
+ "platform": "tinker",
2385
+ "peak_accuracy": 1.0,
2386
+ "last10_accuracy": 0.85
2387
+ },
2388
+ {
2389
+ "experiment": "frontier_gsm8k_nemotron-120b",
2390
+ "model": "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16",
2391
+ "model_short": "nemotron-120b",
2392
+ "task": "gsm8k",
2393
+ "seed": 42,
2394
+ "rank": 16,
2395
+ "lr": 1e-05,
2396
+ "group": 4,
2397
+ "steps": 20,
2398
+ "run_id": "657a920a-9e74-55d2-9354-71a6ec2f1f61:train:0",
2399
+ "checkpoint": "tinker://657a920a-9e74-55d2-9354-71a6ec2f1f61:train:0/sampler_weights/final",
2400
+ "first5_avg": 0.175,
2401
+ "last10_avg": 0.1625,
2402
+ "peak": 0.875,
2403
+ "zero_loss_pct": 70.0,
2404
+ "zero_reward_pct": 55.00000000000001,
2405
+ "reward_trace": [
2406
+ 0.0,
2407
+ 0.0,
2408
+ 0.875,
2409
+ 0.0,
2410
+ 0.0,
2411
+ 0.0,
2412
+ 0.125,
2413
+ 0.5,
2414
+ 0.375,
2415
+ 0.0,
2416
+ 0.125,
2417
+ 0.25,
2418
+ 0.0,
2419
+ 0.0,
2420
+ 0.25,
2421
+ 0.0,
2422
+ 0.0,
2423
+ 0.0,
2424
+ 0.5,
2425
+ 0.5
2426
+ ],
2427
+ "step_log": [
2428
+ {
2429
+ "step": 1,
2430
+ "loss": 0.0,
2431
+ "reward": 0.0,
2432
+ "zvf": 1.0,
2433
+ "gu": 0.0
2434
+ },
2435
+ {
2436
+ "step": 2,
2437
+ "loss": 0.0,
2438
+ "reward": 0.0,
2439
+ "zvf": 1.0,
2440
+ "gu": 0.0
2441
+ },
2442
+ {
2443
+ "step": 3,
2444
+ "loss": -42.757484436035156,
2445
+ "reward": 0.875,
2446
+ "zvf": 0.0,
2447
+ "gu": 1.0
2448
+ },
2449
+ {
2450
+ "step": 4,
2451
+ "loss": 0.0,
2452
+ "reward": 0.0,
2453
+ "zvf": 1.0,
2454
+ "gu": 0.0
2455
+ },
2456
+ {
2457
+ "step": 5,
2458
+ "loss": 0.0,
2459
+ "reward": 0.0,
2460
+ "zvf": 1.0,
2461
+ "gu": 0.0
2462
+ },
2463
+ {
2464
+ "step": 6,
2465
+ "loss": 0.0,
2466
+ "reward": 0.0,
2467
+ "zvf": 1.0,
2468
+ "gu": 0.0
2469
+ },
2470
+ {
2471
+ "step": 7,
2472
+ "loss": 5.0445404052734375,
2473
+ "reward": 0.125,
2474
+ "zvf": 0.0,
2475
+ "gu": 1.0
2476
+ },
2477
+ {
2478
+ "step": 8,
2479
+ "loss": 0.0,
2480
+ "reward": 0.5,
2481
+ "zvf": 0.0,
2482
+ "gu": 1.0
2483
+ },
2484
+ {
2485
+ "step": 9,
2486
+ "loss": -100.07317352294922,
2487
+ "reward": 0.375,
2488
+ "zvf": 0.0,
2489
+ "gu": 1.0
2490
+ },
2491
+ {
2492
+ "step": 10,
2493
+ "loss": 0.0,
2494
+ "reward": 0.0,
2495
+ "zvf": 1.0,
2496
+ "gu": 0.0
2497
+ },
2498
+ {
2499
+ "step": 11,
2500
+ "loss": -14.067096710205078,
2501
+ "reward": 0.125,
2502
+ "zvf": 0.0,
2503
+ "gu": 1.0
2504
+ },
2505
+ {
2506
+ "step": 12,
2507
+ "loss": -76.11570739746094,
2508
+ "reward": 0.25,
2509
+ "zvf": 0.0,
2510
+ "gu": 1.0
2511
+ },
2512
+ {
2513
+ "step": 13,
2514
+ "loss": 0.0,
2515
+ "reward": 0.0,
2516
+ "zvf": 1.0,
2517
+ "gu": 0.0
2518
+ },
2519
+ {
2520
+ "step": 14,
2521
+ "loss": 0.0,
2522
+ "reward": 0.0,
2523
+ "zvf": 1.0,
2524
+ "gu": 0.0
2525
+ },
2526
+ {
2527
+ "step": 15,
2528
+ "loss": 70.30506896972656,
2529
+ "reward": 0.25,
2530
+ "zvf": 0.0,
2531
+ "gu": 1.0
2532
+ },
2533
+ {
2534
+ "step": 16,
2535
+ "loss": 0.0,
2536
+ "reward": 0.0,
2537
+ "zvf": 1.0,
2538
+ "gu": 0.0
2539
+ },
2540
+ {
2541
+ "step": 17,
2542
+ "loss": 0.0,
2543
+ "reward": 0.0,
2544
+ "zvf": 1.0,
2545
+ "gu": 0.0
2546
+ },
2547
+ {
2548
+ "step": 18,
2549
+ "loss": 0.0,
2550
+ "reward": 0.0,
2551
+ "zvf": 1.0,
2552
+ "gu": 0.0
2553
+ },
2554
+ {
2555
+ "step": 19,
2556
+ "loss": 0.0,
2557
+ "reward": 0.5,
2558
+ "zvf": 0.0,
2559
+ "gu": 1.0
2560
+ },
2561
+ {
2562
+ "step": 20,
2563
+ "loss": 0.0,
2564
+ "reward": 0.5,
2565
+ "zvf": 0.0,
2566
+ "gu": 1.0
2567
+ }
2568
+ ],
2569
+ "timestamp": "2026-04-19T03:31:55.848278",
2570
+ "platform": "tinker",
2571
+ "peak_accuracy": 0.875,
2572
+ "last10_accuracy": 0.1625
2573
+ },
2574
+ {
2575
+ "experiment": "frontier_gsm8k_qwen3-235b",
2576
+ "model": "Qwen/Qwen3-235B-A22B-Instruct-2507",
2577
+ "model_short": "qwen3-235b-moe",
2578
+ "task": "gsm8k",
2579
+ "seed": 42,
2580
+ "rank": 16,
2581
+ "lr": 1e-05,
2582
+ "group": 4,
2583
+ "steps": 20,
2584
+ "steps_completed": 15,
2585
+ "partial": true,
2586
+ "run_id": "partial-from-console",
2587
+ "checkpoint": "training_interrupted",
2588
+ "first5_avg": 1.0,
2589
+ "last10_avg": 1.0,
2590
+ "peak": 1.0,
2591
+ "zero_loss_pct": 0,
2592
+ "zero_reward_pct": 0.0,
2593
+ "reward_trace": [
2594
+ 1.0,
2595
+ 1.0,
2596
+ 1.0,
2597
+ 1.0
2598
+ ],
2599
+ "platform": "tinker",
2600
+ "peak_accuracy": 1.0,
2601
+ "last10_accuracy": 1.0
2602
+ },
2603
+ {
2604
+ "experiment": "moe_gsm8k_qwen3-30b-inst",
2605
+ "model": "Qwen/Qwen3-30B-A3B-Instruct-2507",
2606
+ "model_short": "qwen3-30b-moe-inst",
2607
+ "task": "gsm8k",
2608
+ "seed": 42,
2609
+ "rank": 32,
2610
+ "lr": 3e-05,
2611
+ "group": 8,
2612
+ "steps": 30,
2613
+ "steps_completed": 10,
2614
+ "partial": true,
2615
+ "run_id": "partial-from-console",
2616
+ "checkpoint": "training_interrupted",
2617
+ "first5_avg": 1.0,
2618
+ "last10_avg": 1.0,
2619
+ "peak": 1.0,
2620
+ "zero_loss_pct": 0,
2621
+ "zero_reward_pct": 0.0,
2622
+ "reward_trace": [
2623
+ 1.0,
2624
+ 1.0,
2625
+ 1.0
2626
+ ],
2627
+ "platform": "tinker",
2628
+ "peak_accuracy": 1.0,
2629
+ "last10_accuracy": 1.0
2630
+ },
2631
+ {
2632
+ "experiment": "moe_gsm8k_qwen3-30b-moe",
2633
+ "model": "Qwen/Qwen3-30B-A3B",
2634
+ "model_short": "qwen3-30b-moe",
2635
+ "task": "gsm8k",
2636
+ "seed": 42,
2637
+ "rank": 32,
2638
+ "lr": 3e-05,
2639
+ "group": 8,
2640
+ "steps": 30,
2641
+ "steps_completed": 20,
2642
+ "partial": true,
2643
+ "run_id": "partial-from-console",
2644
+ "checkpoint": "training_interrupted",
2645
+ "first5_avg": 0.3252,
2646
+ "last10_avg": 0.3252,
2647
+ "peak": 0.5,
2648
+ "zero_loss_pct": 0,
2649
+ "zero_reward_pct": 0.0,
2650
+ "reward_trace": [
2651
+ 0.188,
2652
+ 0.438,
2653
+ 0.062,
2654
+ 0.5,
2655
+ 0.438
2656
+ ],
2657
+ "platform": "tinker",
2658
+ "peak_accuracy": 0.5,
2659
+ "last10_accuracy": 0.3252
2660
+ },
2661
+ {
2662
+ "experiment": "scale_gsm8k_llama-8b-inst",
2663
+ "model": "meta-llama/Llama-3.1-8B-Instruct",
2664
+ "model_short": "llama-8b-inst",
2665
+ "task": "gsm8k",
2666
+ "seed": 42,
2667
+ "rank": 32,
2668
+ "lr": 3e-05,
2669
+ "group": 8,
2670
+ "steps": 30,
2671
+ "run_id": "488bbb2e-aa35-5431-b4a5-793f68b6bc49:train:0",
2672
+ "checkpoint": "tinker://488bbb2e-aa35-5431-b4a5-793f68b6bc49:train:0/sampler_weights/final",
2673
+ "first5_avg": 0.975,
2674
+ "last10_avg": 0.84375,
2675
+ "peak": 1.0,
2676
+ "zero_loss_pct": 43.333333333333336,
2677
+ "zero_reward_pct": 0.0,
2678
+ "reward_trace": [
2679
+ 1.0,
2680
+ 0.875,
2681
+ 1.0,
2682
+ 1.0,
2683
+ 1.0,
2684
+ 0.625,
2685
+ 1.0,
2686
+ 1.0,
2687
+ 0.875,
2688
+ 0.875,
2689
+ 1.0,
2690
+ 1.0,
2691
+ 0.9375,
2692
+ 0.75,
2693
+ 1.0,
2694
+ 0.375,
2695
+ 0.8125,
2696
+ 0.875,
2697
+ 0.875,
2698
+ 0.75,
2699
+ 0.875,
2700
+ 0.6875,
2701
+ 0.75,
2702
+ 0.4375,
2703
+ 1.0,
2704
+ 1.0,
2705
+ 0.8125,
2706
+ 1.0,
2707
+ 0.875,
2708
+ 1.0
2709
+ ],
2710
+ "step_log": [
2711
+ {
2712
+ "step": 1,
2713
+ "loss": 0.0,
2714
+ "reward": 1.0,
2715
+ "zvf": 1.0,
2716
+ "gu": 0.0
2717
+ },
2718
+ {
2719
+ "step": 2,
2720
+ "loss": 6.620754241943359,
2721
+ "reward": 0.875,
2722
+ "zvf": 0.0,
2723
+ "gu": 1.0
2724
+ },
2725
+ {
2726
+ "step": 3,
2727
+ "loss": 0.0,
2728
+ "reward": 1.0,
2729
+ "zvf": 1.0,
2730
+ "gu": 0.0
2731
+ },
2732
+ {
2733
+ "step": 4,
2734
+ "loss": 0.0,
2735
+ "reward": 1.0,
2736
+ "zvf": 1.0,
2737
+ "gu": 0.0
2738
+ },
2739
+ {
2740
+ "step": 5,
2741
+ "loss": 0.0,
2742
+ "reward": 1.0,
2743
+ "zvf": 1.0,
2744
+ "gu": 0.0
2745
+ },
2746
+ {
2747
+ "step": 6,
2748
+ "loss": 2.9265289306640625,
2749
+ "reward": 0.625,
2750
+ "zvf": 0.0,
2751
+ "gu": 1.0
2752
+ },
2753
+ {
2754
+ "step": 7,
2755
+ "loss": 0.0,
2756
+ "reward": 1.0,
2757
+ "zvf": 1.0,
2758
+ "gu": 0.0
2759
+ },
2760
+ {
2761
+ "step": 8,
2762
+ "loss": 0.0,
2763
+ "reward": 1.0,
2764
+ "zvf": 1.0,
2765
+ "gu": 0.0
2766
+ },
2767
+ {
2768
+ "step": 9,
2769
+ "loss": -0.9360866546630859,
2770
+ "reward": 0.875,
2771
+ "zvf": 0.0,
2772
+ "gu": 1.0
2773
+ },
2774
+ {
2775
+ "step": 10,
2776
+ "loss": -2.864093780517578,
2777
+ "reward": 0.875,
2778
+ "zvf": 0.0,
2779
+ "gu": 1.0
2780
+ },
2781
+ {
2782
+ "step": 11,
2783
+ "loss": 0.0,
2784
+ "reward": 1.0,
2785
+ "zvf": 1.0,
2786
+ "gu": 0.0
2787
+ },
2788
+ {
2789
+ "step": 12,
2790
+ "loss": 0.0,
2791
+ "reward": 1.0,
2792
+ "zvf": 1.0,
2793
+ "gu": 0.0
2794
+ },
2795
+ {
2796
+ "step": 13,
2797
+ "loss": -22.385391235351562,
2798
+ "reward": 0.9375,
2799
+ "zvf": 0.0,
2800
+ "gu": 1.0
2801
+ },
2802
+ {
2803
+ "step": 14,
2804
+ "loss": -5.675556182861328,
2805
+ "reward": 0.75,
2806
+ "zvf": 0.0,
2807
+ "gu": 1.0
2808
+ },
2809
+ {
2810
+ "step": 15,
2811
+ "loss": 0.0,
2812
+ "reward": 1.0,
2813
+ "zvf": 1.0,
2814
+ "gu": 0.0
2815
+ },
2816
+ {
2817
+ "step": 16,
2818
+ "loss": 7.295082092285156,
2819
+ "reward": 0.375,
2820
+ "zvf": 0.0,
2821
+ "gu": 1.0
2822
+ },
2823
+ {
2824
+ "step": 17,
2825
+ "loss": 2.713848114013672,
2826
+ "reward": 0.8125,
2827
+ "zvf": 0.0,
2828
+ "gu": 1.0
2829
+ },
2830
+ {
2831
+ "step": 18,
2832
+ "loss": -18.252927780151367,
2833
+ "reward": 0.875,
2834
+ "zvf": 0.0,
2835
+ "gu": 1.0
2836
+ },
2837
+ {
2838
+ "step": 19,
2839
+ "loss": -5.822566986083984,
2840
+ "reward": 0.875,
2841
+ "zvf": 0.0,
2842
+ "gu": 1.0
2843
+ },
2844
+ {
2845
+ "step": 20,
2846
+ "loss": 8.464057922363281,
2847
+ "reward": 0.75,
2848
+ "zvf": 0.0,
2849
+ "gu": 1.0
2850
+ },
2851
+ {
2852
+ "step": 21,
2853
+ "loss": 13.122426986694336,
2854
+ "reward": 0.875,
2855
+ "zvf": 0.0,
2856
+ "gu": 1.0
2857
+ },
2858
+ {
2859
+ "step": 22,
2860
+ "loss": -19.496044158935547,
2861
+ "reward": 0.6875,
2862
+ "zvf": 0.0,
2863
+ "gu": 1.0
2864
+ },
2865
+ {
2866
+ "step": 23,
2867
+ "loss": 0.4438667297363281,
2868
+ "reward": 0.75,
2869
+ "zvf": 0.0,
2870
+ "gu": 1.0
2871
+ },
2872
+ {
2873
+ "step": 24,
2874
+ "loss": 4.036277770996094,
2875
+ "reward": 0.4375,
2876
+ "zvf": 0.0,
2877
+ "gu": 1.0
2878
+ },
2879
+ {
2880
+ "step": 25,
2881
+ "loss": 0.0,
2882
+ "reward": 1.0,
2883
+ "zvf": 1.0,
2884
+ "gu": 0.0
2885
+ },
2886
+ {
2887
+ "step": 26,
2888
+ "loss": 0.0,
2889
+ "reward": 1.0,
2890
+ "zvf": 1.0,
2891
+ "gu": 0.0
2892
+ },
2893
+ {
2894
+ "step": 27,
2895
+ "loss": 3.3459701538085938,
2896
+ "reward": 0.8125,
2897
+ "zvf": 0.0,
2898
+ "gu": 1.0
2899
+ },
2900
+ {
2901
+ "step": 28,
2902
+ "loss": 0.0,
2903
+ "reward": 1.0,
2904
+ "zvf": 1.0,
2905
+ "gu": 0.0
2906
+ },
2907
+ {
2908
+ "step": 29,
2909
+ "loss": 12.346473693847656,
2910
+ "reward": 0.875,
2911
+ "zvf": 0.0,
2912
+ "gu": 1.0
2913
+ },
2914
+ {
2915
+ "step": 30,
2916
+ "loss": 0.0,
2917
+ "reward": 1.0,
2918
+ "zvf": 1.0,
2919
+ "gu": 0.0
2920
+ }
2921
+ ],
2922
+ "timestamp": "2026-04-19T04:11:44.481966",
2923
+ "platform": "tinker",
2924
+ "peak_accuracy": 1.0,
2925
+ "last10_accuracy": 0.84375
2926
+ },
2927
+ {
2928
+ "experiment": "scale_gsm8k_qwen3-32b",
2929
+ "model": "Qwen/Qwen3-32B",
2930
+ "model_short": "qwen3-32b",
2931
+ "task": "gsm8k",
2932
+ "seed": 42,
2933
+ "rank": 32,
2934
+ "lr": 3e-05,
2935
+ "group": 8,
2936
+ "steps": 30,
2937
+ "steps_completed": 10,
2938
+ "partial": true,
2939
+ "run_id": "partial-from-console",
2940
+ "checkpoint": "training_interrupted",
2941
+ "first5_avg": 0.24966666666666668,
2942
+ "last10_avg": 0.24966666666666668,
2943
+ "peak": 0.312,
2944
+ "zero_loss_pct": 0,
2945
+ "zero_reward_pct": 0.0,
2946
+ "reward_trace": [
2947
+ 0.312,
2948
+ 0.312,
2949
+ 0.125
2950
+ ],
2951
+ "platform": "tinker",
2952
+ "peak_accuracy": 0.312,
2953
+ "last10_accuracy": 0.24966666666666668
2954
+ },
2955
+ {
2956
+ "experiment": "scale_gsm8k_qwen3-8b",
2957
+ "model": "Qwen/Qwen3-8B",
2958
+ "model_short": "qwen3-8b",
2959
+ "task": "gsm8k",
2960
+ "seed": 42,
2961
+ "rank": 32,
2962
+ "lr": 3e-05,
2963
+ "group": 8,
2964
+ "steps": 30,
2965
+ "run_id": "3154dee5-4fd7-59b6-ba3b-721eef675bfc:train:0",
2966
+ "checkpoint": "tinker://3154dee5-4fd7-59b6-ba3b-721eef675bfc:train:0/sampler_weights/final",
2967
+ "first5_avg": 0.3125,
2968
+ "last10_avg": 0.34375,
2969
+ "peak": 0.625,
2970
+ "zero_loss_pct": 6.666666666666667,
2971
+ "zero_reward_pct": 6.666666666666667,
2972
+ "reward_trace": [
2973
+ 0.375,
2974
+ 0.0625,
2975
+ 0.5625,
2976
+ 0.1875,
2977
+ 0.375,
2978
+ 0.3125,
2979
+ 0.125,
2980
+ 0.125,
2981
+ 0.1875,
2982
+ 0.125,
2983
+ 0.25,
2984
+ 0.4375,
2985
+ 0.625,
2986
+ 0.0,
2987
+ 0.4375,
2988
+ 0.0625,
2989
+ 0.1875,
2990
+ 0.0,
2991
+ 0.375,
2992
+ 0.3125,
2993
+ 0.3125,
2994
+ 0.1875,
2995
+ 0.375,
2996
+ 0.375,
2997
+ 0.5,
2998
+ 0.25,
2999
+ 0.4375,
3000
+ 0.1875,
3001
+ 0.625,
3002
+ 0.1875
3003
+ ],
3004
+ "step_log": [
3005
+ {
3006
+ "step": 1,
3007
+ "loss": -3.936065673828125,
3008
+ "reward": 0.375,
3009
+ "zvf": 0.0,
3010
+ "gu": 1.0
3011
+ },
3012
+ {
3013
+ "step": 2,
3014
+ "loss": -2.969602584838867,
3015
+ "reward": 0.0625,
3016
+ "zvf": 0.0,
3017
+ "gu": 1.0
3018
+ },
3019
+ {
3020
+ "step": 3,
3021
+ "loss": -4.0252532958984375,
3022
+ "reward": 0.5625,
3023
+ "zvf": 0.0,
3024
+ "gu": 1.0
3025
+ },
3026
+ {
3027
+ "step": 4,
3028
+ "loss": -3.1384735107421875,
3029
+ "reward": 0.1875,
3030
+ "zvf": 0.0,
3031
+ "gu": 1.0
3032
+ },
3033
+ {
3034
+ "step": 5,
3035
+ "loss": -5.50750732421875,
3036
+ "reward": 0.375,
3037
+ "zvf": 0.0,
3038
+ "gu": 1.0
3039
+ },
3040
+ {
3041
+ "step": 6,
3042
+ "loss": -2.173931121826172,
3043
+ "reward": 0.3125,
3044
+ "zvf": 0.0,
3045
+ "gu": 1.0
3046
+ },
3047
+ {
3048
+ "step": 7,
3049
+ "loss": -2.5752410888671875,
3050
+ "reward": 0.125,
3051
+ "zvf": 0.0,
3052
+ "gu": 1.0
3053
+ },
3054
+ {
3055
+ "step": 8,
3056
+ "loss": -2.2352685928344727,
3057
+ "reward": 0.125,
3058
+ "zvf": 0.0,
3059
+ "gu": 1.0
3060
+ },
3061
+ {
3062
+ "step": 9,
3063
+ "loss": -4.3625383377075195,
3064
+ "reward": 0.1875,
3065
+ "zvf": 0.0,
3066
+ "gu": 1.0
3067
+ },
3068
+ {
3069
+ "step": 10,
3070
+ "loss": -1.4611568450927734,
3071
+ "reward": 0.125,
3072
+ "zvf": 0.0,
3073
+ "gu": 1.0
3074
+ },
3075
+ {
3076
+ "step": 11,
3077
+ "loss": -8.721845626831055,
3078
+ "reward": 0.25,
3079
+ "zvf": 0.0,
3080
+ "gu": 1.0
3081
+ },
3082
+ {
3083
+ "step": 12,
3084
+ "loss": 1.46728515625,
3085
+ "reward": 0.4375,
3086
+ "zvf": 0.0,
3087
+ "gu": 1.0
3088
+ },
3089
+ {
3090
+ "step": 13,
3091
+ "loss": -6.392450332641602,
3092
+ "reward": 0.625,
3093
+ "zvf": 0.0,
3094
+ "gu": 1.0
3095
+ },
3096
+ {
3097
+ "step": 14,
3098
+ "loss": 0.0,
3099
+ "reward": 0.0,
3100
+ "zvf": 1.0,
3101
+ "gu": 0.0
3102
+ },
3103
+ {
3104
+ "step": 15,
3105
+ "loss": -2.3406524658203125,
3106
+ "reward": 0.4375,
3107
+ "zvf": 0.0,
3108
+ "gu": 1.0
3109
+ },
3110
+ {
3111
+ "step": 16,
3112
+ "loss": 1.5611343383789062,
3113
+ "reward": 0.0625,
3114
+ "zvf": 0.0,
3115
+ "gu": 1.0
3116
+ },
3117
+ {
3118
+ "step": 17,
3119
+ "loss": -2.0525875091552734,
3120
+ "reward": 0.1875,
3121
+ "zvf": 0.0,
3122
+ "gu": 1.0
3123
+ },
3124
+ {
3125
+ "step": 18,
3126
+ "loss": 0.0,
3127
+ "reward": 0.0,
3128
+ "zvf": 1.0,
3129
+ "gu": 0.0
3130
+ },
3131
+ {
3132
+ "step": 19,
3133
+ "loss": -7.680427551269531,
3134
+ "reward": 0.375,
3135
+ "zvf": 0.0,
3136
+ "gu": 1.0
3137
+ },
3138
+ {
3139
+ "step": 20,
3140
+ "loss": -3.0284852981567383,
3141
+ "reward": 0.3125,
3142
+ "zvf": 0.0,
3143
+ "gu": 1.0
3144
+ },
3145
+ {
3146
+ "step": 21,
3147
+ "loss": -1.5452747344970703,
3148
+ "reward": 0.3125,
3149
+ "zvf": 0.0,
3150
+ "gu": 1.0
3151
+ },
3152
+ {
3153
+ "step": 22,
3154
+ "loss": -2.5575408935546875,
3155
+ "reward": 0.1875,
3156
+ "zvf": 0.0,
3157
+ "gu": 1.0
3158
+ },
3159
+ {
3160
+ "step": 23,
3161
+ "loss": -4.197565078735352,
3162
+ "reward": 0.375,
3163
+ "zvf": 0.0,
3164
+ "gu": 1.0
3165
+ },
3166
+ {
3167
+ "step": 24,
3168
+ "loss": -3.269200325012207,
3169
+ "reward": 0.375,
3170
+ "zvf": 0.0,
3171
+ "gu": 1.0
3172
+ },
3173
+ {
3174
+ "step": 25,
3175
+ "loss": -6.419197082519531,
3176
+ "reward": 0.5,
3177
+ "zvf": 0.0,
3178
+ "gu": 1.0
3179
+ },
3180
+ {
3181
+ "step": 26,
3182
+ "loss": -1.5014839172363281,
3183
+ "reward": 0.25,
3184
+ "zvf": 0.0,
3185
+ "gu": 1.0
3186
+ },
3187
+ {
3188
+ "step": 27,
3189
+ "loss": -4.115062713623047,
3190
+ "reward": 0.4375,
3191
+ "zvf": 0.0,
3192
+ "gu": 1.0
3193
+ },
3194
+ {
3195
+ "step": 28,
3196
+ "loss": -2.7128124237060547,
3197
+ "reward": 0.1875,
3198
+ "zvf": 0.0,
3199
+ "gu": 1.0
3200
+ },
3201
+ {
3202
+ "step": 29,
3203
+ "loss": -2.6960220336914062,
3204
+ "reward": 0.625,
3205
+ "zvf": 0.0,
3206
+ "gu": 1.0
3207
+ },
3208
+ {
3209
+ "step": 30,
3210
+ "loss": -0.14708518981933594,
3211
+ "reward": 0.1875,
3212
+ "zvf": 0.0,
3213
+ "gu": 1.0
3214
+ }
3215
+ ],
3216
+ "timestamp": "2026-04-18T16:31:17.561535",
3217
+ "platform": "tinker",
3218
+ "peak_accuracy": 0.625,
3219
+ "last10_accuracy": 0.34375
3220
+ },
3221
+ {
3222
+ "experiment": "scale_gsm8k_qwen3.5-27b",
3223
+ "model": "Qwen/Qwen3.5-27B",
3224
+ "model_short": "qwen3.5-27b",
3225
+ "task": "gsm8k",
3226
+ "seed": 42,
3227
+ "rank": 32,
3228
+ "lr": 3e-05,
3229
+ "group": 8,
3230
+ "steps": 30,
3231
+ "steps_completed": 10,
3232
+ "partial": true,
3233
+ "run_id": "partial-from-console",
3234
+ "checkpoint": "training_interrupted",
3235
+ "first5_avg": 0.43733333333333335,
3236
+ "last10_avg": 0.43733333333333335,
3237
+ "peak": 0.75,
3238
+ "zero_loss_pct": 0,
3239
+ "zero_reward_pct": 33.33333333333333,
3240
+ "reward_trace": [
3241
+ 0.562,
3242
+ 0.75,
3243
+ 0.0
3244
+ ],
3245
+ "platform": "tinker",
3246
+ "peak_accuracy": 0.75,
3247
+ "last10_accuracy": 0.43733333333333335
3248
+ },
3249
+ {
3250
+ "experiment": "scale_gsm8k_qwen3.5-4b",
3251
+ "model": "Qwen/Qwen3.5-4B",
3252
+ "model_short": "qwen3.5-4b",
3253
+ "task": "gsm8k",
3254
+ "seed": 42,
3255
+ "rank": 32,
3256
+ "lr": 3e-05,
3257
+ "group": 8,
3258
+ "steps": 30,
3259
+ "run_id": "200e4983-3e9d-5176-87d0-c55dd3c73539:train:0",
3260
+ "checkpoint": "tinker://200e4983-3e9d-5176-87d0-c55dd3c73539:train:0/sampler_weights/final",
3261
+ "first5_avg": 0.9625,
3262
+ "last10_avg": 0.85,
3263
+ "peak": 1.0,
3264
+ "zero_loss_pct": 56.666666666666664,
3265
+ "zero_reward_pct": 0.0,
3266
+ "reward_trace": [
3267
+ 1.0,
3268
+ 0.8125,
3269
+ 1.0,
3270
+ 1.0,
3271
+ 1.0,
3272
+ 0.5,
3273
+ 0.5,
3274
+ 0.875,
3275
+ 0.625,
3276
+ 0.8125,
3277
+ 1.0,
3278
+ 1.0,
3279
+ 1.0,
3280
+ 0.625,
3281
+ 0.875,
3282
+ 0.25,
3283
+ 1.0,
3284
+ 0.875,
3285
+ 0.6875,
3286
+ 0.5625,
3287
+ 0.9375,
3288
+ 0.5,
3289
+ 0.9375,
3290
+ 0.5,
3291
+ 1.0,
3292
+ 1.0,
3293
+ 0.625,
3294
+ 1.0,
3295
+ 1.0,
3296
+ 1.0
3297
+ ],
3298
+ "step_log": [
3299
+ {
3300
+ "step": 1,
3301
+ "loss": 0.0,
3302
+ "reward": 1.0,
3303
+ "zvf": 1.0,
3304
+ "gu": 0.0
3305
+ },
3306
+ {
3307
+ "step": 2,
3308
+ "loss": -2.847966194152832,
3309
+ "reward": 0.8125,
3310
+ "zvf": 0.0,
3311
+ "gu": 1.0
3312
+ },
3313
+ {
3314
+ "step": 3,
3315
+ "loss": 0.0,
3316
+ "reward": 1.0,
3317
+ "zvf": 1.0,
3318
+ "gu": 0.0
3319
+ },
3320
+ {
3321
+ "step": 4,
3322
+ "loss": 0.0,
3323
+ "reward": 1.0,
3324
+ "zvf": 1.0,
3325
+ "gu": 0.0
3326
+ },
3327
+ {
3328
+ "step": 5,
3329
+ "loss": 0.0,
3330
+ "reward": 1.0,
3331
+ "zvf": 1.0,
3332
+ "gu": 0.0
3333
+ },
3334
+ {
3335
+ "step": 6,
3336
+ "loss": 0.0,
3337
+ "reward": 0.5,
3338
+ "zvf": 0.0,
3339
+ "gu": 1.0
3340
+ },
3341
+ {
3342
+ "step": 7,
3343
+ "loss": 0.0,
3344
+ "reward": 0.5,
3345
+ "zvf": 0.0,
3346
+ "gu": 1.0
3347
+ },
3348
+ {
3349
+ "step": 8,
3350
+ "loss": -1.0574636459350586,
3351
+ "reward": 0.875,
3352
+ "zvf": 0.0,
3353
+ "gu": 1.0
3354
+ },
3355
+ {
3356
+ "step": 9,
3357
+ "loss": -1.7392759323120117,
3358
+ "reward": 0.625,
3359
+ "zvf": 0.0,
3360
+ "gu": 1.0
3361
+ },
3362
+ {
3363
+ "step": 10,
3364
+ "loss": -23.38077163696289,
3365
+ "reward": 0.8125,
3366
+ "zvf": 0.0,
3367
+ "gu": 1.0
3368
+ },
3369
+ {
3370
+ "step": 11,
3371
+ "loss": 0.0,
3372
+ "reward": 1.0,
3373
+ "zvf": 1.0,
3374
+ "gu": 0.0
3375
+ },
3376
+ {
3377
+ "step": 12,
3378
+ "loss": 0.0,
3379
+ "reward": 1.0,
3380
+ "zvf": 1.0,
3381
+ "gu": 0.0
3382
+ },
3383
+ {
3384
+ "step": 13,
3385
+ "loss": 0.0,
3386
+ "reward": 1.0,
3387
+ "zvf": 1.0,
3388
+ "gu": 0.0
3389
+ },
3390
+ {
3391
+ "step": 14,
3392
+ "loss": -5.09890079498291,
3393
+ "reward": 0.625,
3394
+ "zvf": 0.0,
3395
+ "gu": 1.0
3396
+ },
3397
+ {
3398
+ "step": 15,
3399
+ "loss": -17.38865089416504,
3400
+ "reward": 0.875,
3401
+ "zvf": 0.0,
3402
+ "gu": 1.0
3403
+ },
3404
+ {
3405
+ "step": 16,
3406
+ "loss": -10.52424430847168,
3407
+ "reward": 0.25,
3408
+ "zvf": 0.0,
3409
+ "gu": 1.0
3410
+ },
3411
+ {
3412
+ "step": 17,
3413
+ "loss": 0.0,
3414
+ "reward": 1.0,
3415
+ "zvf": 1.0,
3416
+ "gu": 0.0
3417
+ },
3418
+ {
3419
+ "step": 18,
3420
+ "loss": -12.190226554870605,
3421
+ "reward": 0.875,
3422
+ "zvf": 0.0,
3423
+ "gu": 1.0
3424
+ },
3425
+ {
3426
+ "step": 19,
3427
+ "loss": -6.337299346923828,
3428
+ "reward": 0.6875,
3429
+ "zvf": 0.0,
3430
+ "gu": 1.0
3431
+ },
3432
+ {
3433
+ "step": 20,
3434
+ "loss": -4.9612884521484375,
3435
+ "reward": 0.5625,
3436
+ "zvf": 0.0,
3437
+ "gu": 1.0
3438
+ },
3439
+ {
3440
+ "step": 21,
3441
+ "loss": -10.336174011230469,
3442
+ "reward": 0.9375,
3443
+ "zvf": 0.0,
3444
+ "gu": 1.0
3445
+ },
3446
+ {
3447
+ "step": 22,
3448
+ "loss": 0.0,
3449
+ "reward": 0.5,
3450
+ "zvf": 0.0,
3451
+ "gu": 1.0
3452
+ },
3453
+ {
3454
+ "step": 23,
3455
+ "loss": -5.058204650878906,
3456
+ "reward": 0.9375,
3457
+ "zvf": 0.0,
3458
+ "gu": 1.0
3459
+ },
3460
+ {
3461
+ "step": 24,
3462
+ "loss": 0.0,
3463
+ "reward": 0.5,
3464
+ "zvf": 0.0,
3465
+ "gu": 1.0
3466
+ },
3467
+ {
3468
+ "step": 25,
3469
+ "loss": 0.0,
3470
+ "reward": 1.0,
3471
+ "zvf": 1.0,
3472
+ "gu": 0.0
3473
+ },
3474
+ {
3475
+ "step": 26,
3476
+ "loss": 0.0,
3477
+ "reward": 1.0,
3478
+ "zvf": 1.0,
3479
+ "gu": 0.0
3480
+ },
3481
+ {
3482
+ "step": 27,
3483
+ "loss": -7.397833824157715,
3484
+ "reward": 0.625,
3485
+ "zvf": 0.0,
3486
+ "gu": 1.0
3487
+ },
3488
+ {
3489
+ "step": 28,
3490
+ "loss": 0.0,
3491
+ "reward": 1.0,
3492
+ "zvf": 1.0,
3493
+ "gu": 0.0
3494
+ },
3495
+ {
3496
+ "step": 29,
3497
+ "loss": 0.0,
3498
+ "reward": 1.0,
3499
+ "zvf": 1.0,
3500
+ "gu": 0.0
3501
+ },
3502
+ {
3503
+ "step": 30,
3504
+ "loss": 0.0,
3505
+ "reward": 1.0,
3506
+ "zvf": 1.0,
3507
+ "gu": 0.0
3508
+ }
3509
+ ],
3510
+ "timestamp": "2026-04-19T03:34:22.495977",
3511
+ "platform": "tinker",
3512
+ "peak_accuracy": 1.0,
3513
+ "last10_accuracy": 0.85
3514
+ },
3515
+ {
3516
+ "experiment": "kl_track_Qwen3-8B_s42",
3517
+ "model": "Qwen/Qwen3-8B",
3518
+ "model_short": "qwen3-8b",
3519
+ "task": "kl_tracking",
3520
+ "seed": 42,
3521
+ "steps": 30,
3522
+ "final_kl": 60.75,
3523
+ "final_entropy": 0.742,
3524
+ "platform": "modal_h100"
3525
+ },
3526
+ {
3527
+ "experiment": "ppo_gsm8k_qwen3-8b",
3528
+ "model": "Qwen/Qwen3-8B",
3529
+ "model_short": "qwen3-8b",
3530
+ "task": "gsm8k",
3531
+ "seed": 42,
3532
+ "steps": 30,
3533
+ "peak": 0.75,
3534
+ "last10_avg": 0.225,
3535
+ "platform": "modal_h100",
3536
+ "algorithm": "PPO"
3537
+ },
3538
+ {
3539
+ "experiment": "ppo_gsm8k_llama-8b",
3540
+ "model": "meta-llama/Llama-3.1-8B-Instruct",
3541
+ "model_short": "llama-8b-inst",
3542
+ "task": "gsm8k",
3543
+ "seed": 42,
3544
+ "steps": 30,
3545
+ "peak": 1.0,
3546
+ "last10_avg": 0.975,
3547
+ "platform": "modal_h100",
3548
+ "algorithm": "PPO"
3549
+ },
3550
+ {
3551
+ "experiment_id": "campaign_v2_w1_qwen3-8b-base",
3552
+ "model": "Qwen/Qwen3-8B-Base",
3553
+ "method": "GRPO",
3554
+ "platform": "Tinker",
3555
+ "task": "gsm8k",
3556
+ "seed": 42,
3557
+ "group_size": 8,
3558
+ "lr": 1e-05,
3559
+ "steps": 30,
3560
+ "peak_reward": 1.0,
3561
+ "last_10_avg": 0.85625,
3562
+ "reward_trace": [
3563
+ 0.6875,
3564
+ 0.9375,
3565
+ 1.0,
3566
+ 0.875,
3567
+ 0.625,
3568
+ 0.625,
3569
+ 1.0,
3570
+ 0.875,
3571
+ 0.9375,
3572
+ 0.8125,
3573
+ 1.0,
3574
+ 0.9375,
3575
+ 0.9375,
3576
+ 0.625,
3577
+ 0.875,
3578
+ 0.9375,
3579
+ 1.0,
3580
+ 1.0,
3581
+ 1.0,
3582
+ 1.0,
3583
+ 0.8125,
3584
+ 1.0,
3585
+ 0.5625,
3586
+ 0.9375,
3587
+ 0.9375,
3588
+ 1.0,
3589
+ 0.8125,
3590
+ 0.9375,
3591
+ 0.875,
3592
+ 0.6875
3593
+ ],
3594
+ "status": "completed",
3595
+ "wave": "1-frontier",
3596
+ "source": "bitter_lesson_campaign_v2"
3597
+ },
3598
+ {
3599
+ "experiment_id": "modal_trl_trl_qwen3_8b",
3600
+ "model": "Qwen/Qwen3-8B",
3601
+ "method": "TRL-GRPO",
3602
+ "platform": "Modal-H100",
3603
+ "task": "gsm8k",
3604
+ "seed": 42,
3605
+ "group_size": 8,
3606
+ "lr": 1e-05,
3607
+ "steps": 30,
3608
+ "peak_reward": 0.375,
3609
+ "last_10_avg": 0.05,
3610
+ "reward_trace": [
3611
+ 0.0,
3612
+ 0.0,
3613
+ 0.0,
3614
+ 0.0,
3615
+ 0.0,
3616
+ 0.125,
3617
+ 0.125,
3618
+ 0.0,
3619
+ 0.0,
3620
+ 0.0,
3621
+ 0.0,
3622
+ 0.0,
3623
+ 0.125,
3624
+ 0.0,
3625
+ 0.0,
3626
+ 0.0,
3627
+ 0.0,
3628
+ 0.0,
3629
+ 0.0,
3630
+ 0.0,
3631
+ 0.0,
3632
+ 0.0,
3633
+ 0.0,
3634
+ 0.375,
3635
+ 0.0,
3636
+ 0.125,
3637
+ 0.0,
3638
+ 0.0,
3639
+ 0.0,
3640
+ 0.0
3641
+ ],
3642
+ "duration_s": 735.7421374320984,
3643
+ "status": "completed",
3644
+ "source": "modal_trl_grpo_campaign"
3645
+ },
3646
+ {
3647
+ "experiment_id": "modal_trl_trl_llama32_3b",
3648
+ "model": "meta-llama/Llama-3.2-3B-Instruct",
3649
+ "method": "TRL-GRPO",
3650
+ "platform": "Modal-H100",
3651
+ "task": "gsm8k",
3652
+ "seed": 42,
3653
+ "group_size": 8,
3654
+ "lr": 1e-05,
3655
+ "steps": 30,
3656
+ "peak_reward": 1.0,
3657
+ "last_10_avg": 0.7125,
3658
+ "reward_trace": [
3659
+ 0.625,
3660
+ 0.5,
3661
+ 0.625,
3662
+ 0.75,
3663
+ 0.875,
3664
+ 1.0,
3665
+ 1.0,
3666
+ 0.625,
3667
+ 1.0,
3668
+ 0.875,
3669
+ 0.5,
3670
+ 0.875,
3671
+ 0.75,
3672
+ 0.875,
3673
+ 0.375,
3674
+ 1.0,
3675
+ 0.875,
3676
+ 0.75,
3677
+ 1.0,
3678
+ 0.25,
3679
+ 0.75,
3680
+ 0.5,
3681
+ 0.75,
3682
+ 0.875,
3683
+ 0.125,
3684
+ 1.0,
3685
+ 0.625,
3686
+ 1.0,
3687
+ 0.625,
3688
+ 0.875
3689
+ ],
3690
+ "duration_s": 222.11492276191711,
3691
+ "status": "completed",
3692
+ "source": "modal_trl_grpo_campaign"
3693
+ },
3694
+ {
3695
+ "experiment_id": "modal_trl_trl_llama32_1b",
3696
+ "model": "meta-llama/Llama-3.2-1B-Instruct",
3697
+ "method": "TRL-GRPO",
3698
+ "platform": "Modal-H100",
3699
+ "task": "gsm8k",
3700
+ "seed": 42,
3701
+ "group_size": 8,
3702
+ "lr": 1e-05,
3703
+ "steps": 30,
3704
+ "peak_reward": 0.875,
3705
+ "last_10_avg": 0.3625,
3706
+ "reward_trace": [
3707
+ 0.125,
3708
+ 0.0,
3709
+ 0.0,
3710
+ 0.125,
3711
+ 0.625,
3712
+ 0.375,
3713
+ 0.125,
3714
+ 0.125,
3715
+ 0.0,
3716
+ 0.125,
3717
+ 0.125,
3718
+ 0.0,
3719
+ 0.0,
3720
+ 0.25,
3721
+ 0.0,
3722
+ 0.375,
3723
+ 0.125,
3724
+ 0.0,
3725
+ 0.375,
3726
+ 0.0,
3727
+ 0.0,
3728
+ 0.625,
3729
+ 0.625,
3730
+ 0.125,
3731
+ 0.125,
3732
+ 0.375,
3733
+ 0.5,
3734
+ 0.875,
3735
+ 0.25,
3736
+ 0.125
3737
+ ],
3738
+ "duration_s": 162.25860285758972,
3739
+ "status": "completed",
3740
+ "source": "modal_trl_grpo_campaign"
3741
+ },
3742
+ {
3743
+ "experiment_id": "campaign_v2_w1_deepseek-v31-base",
3744
+ "model": "deepseek-ai/DeepSeek-V3.1-Base",
3745
+ "method": "GRPO",
3746
+ "platform": "Tinker",
3747
+ "task": "gsm8k",
3748
+ "seed": 42,
3749
+ "group_size": 8,
3750
+ "lr": 1e-05,
3751
+ "steps_completed": 26,
3752
+ "peak_reward": 0.812,
3753
+ "last10_avg": 0.5728333333333334,
3754
+ "first5_avg": 0.525,
3755
+ "reward_trace": [
3756
+ 0.625,
3757
+ 0.312,
3758
+ 0.688,
3759
+ 0.562,
3760
+ 0.438,
3761
+ 0.812
3762
+ ],
3763
+ "status": "completed",
3764
+ "source": "bitter_lesson_campaign_v2"
3765
+ },
3766
+ {
3767
+ "experiment_id": "campaign_v2_w1_gpt-oss-120b",
3768
+ "model": "openai/gpt-oss-120b",
3769
+ "method": "GRPO",
3770
+ "platform": "Tinker",
3771
+ "task": "gsm8k",
3772
+ "seed": 42,
3773
+ "group_size": 8,
3774
+ "lr": 1e-05,
3775
+ "steps_completed": 26,
3776
+ "peak_reward": 1.0,
3777
+ "last10_avg": 0.8753333333333333,
3778
+ "first5_avg": 0.9128000000000001,
3779
+ "reward_trace": [
3780
+ 0.938,
3781
+ 0.938,
3782
+ 0.75,
3783
+ 1.0,
3784
+ 0.938,
3785
+ 0.688
3786
+ ],
3787
+ "status": "completed",
3788
+ "source": "bitter_lesson_campaign_v2"
3789
+ },
3790
+ {
3791
+ "experiment_id": "campaign_v2_w1_kimi-k2-thinking",
3792
+ "model": "moonshotai/Kimi-K2-Thinking",
3793
+ "method": "GRPO",
3794
+ "platform": "Tinker",
3795
+ "task": "gsm8k",
3796
+ "seed": 42,
3797
+ "group_size": 8,
3798
+ "lr": 1e-05,
3799
+ "steps_completed": 26,
3800
+ "peak_reward": 1.0,
3801
+ "last10_avg": 0.9583333333333334,
3802
+ "first5_avg": 0.9875999999999999,
3803
+ "reward_trace": [
3804
+ 1.0,
3805
+ 1.0,
3806
+ 1.0,
3807
+ 0.938,
3808
+ 1.0,
3809
+ 0.812
3810
+ ],
3811
+ "status": "completed",
3812
+ "source": "bitter_lesson_campaign_v2"
3813
+ },
3814
+ {
3815
+ "experiment_id": "campaign_v2_w1_kimi-k25",
3816
+ "model": "moonshotai/Kimi-K2.5",
3817
+ "method": "GRPO",
3818
+ "platform": "Tinker",
3819
+ "task": "gsm8k",
3820
+ "seed": 42,
3821
+ "group_size": 8,
3822
+ "lr": 1e-05,
3823
+ "steps_completed": 26,
3824
+ "peak_reward": 1.0,
3825
+ "last10_avg": 0.625,
3826
+ "first5_avg": 0.55,
3827
+ "reward_trace": [
3828
+ 0.125,
3829
+ 1.0,
3830
+ 0.438,
3831
+ 0.562,
3832
+ 0.625,
3833
+ 1.0
3834
+ ],
3835
+ "status": "completed",
3836
+ "source": "bitter_lesson_campaign_v2"
3837
+ },
3838
+ {
3839
+ "experiment_id": "campaign_v2_w1_llama31-70b-base",
3840
+ "model": "meta-llama/Llama-3.1-70B",
3841
+ "method": "GRPO",
3842
+ "platform": "Tinker",
3843
+ "task": "gsm8k",
3844
+ "seed": 42,
3845
+ "group_size": 8,
3846
+ "lr": 1e-05,
3847
+ "steps_completed": 26,
3848
+ "peak_reward": 0.562,
3849
+ "last10_avg": 0.36433333333333334,
3850
+ "first5_avg": 0.3622,
3851
+ "reward_trace": [
3852
+ 0.312,
3853
+ 0.562,
3854
+ 0.125,
3855
+ 0.562,
3856
+ 0.25,
3857
+ 0.375
3858
+ ],
3859
+ "status": "completed",
3860
+ "source": "bitter_lesson_campaign_v2"
3861
+ },
3862
+ {
3863
+ "experiment_id": "campaign_v2_w1_llama31-8b-base",
3864
+ "model": "meta-llama/Llama-3.1-8B",
3865
+ "method": "GRPO",
3866
+ "platform": "Tinker",
3867
+ "task": "gsm8k",
3868
+ "seed": 42,
3869
+ "group_size": 8,
3870
+ "lr": 1e-05,
3871
+ "steps_completed": 26,
3872
+ "peak_reward": 0.188,
3873
+ "last10_avg": 0.11466666666666665,
3874
+ "first5_avg": 0.1252,
3875
+ "reward_trace": [
3876
+ 0.188,
3877
+ 0.188,
3878
+ 0.0,
3879
+ 0.188,
3880
+ 0.062,
3881
+ 0.062
3882
+ ],
3883
+ "status": "completed",
3884
+ "source": "bitter_lesson_campaign_v2"
3885
+ },
3886
+ {
3887
+ "experiment_id": "campaign_v2_w1_qwen35-397b",
3888
+ "model": "Qwen/Qwen3.5-397B-A17B",
3889
+ "method": "GRPO",
3890
+ "platform": "Tinker",
3891
+ "task": "gsm8k",
3892
+ "seed": 42,
3893
+ "group_size": 8,
3894
+ "lr": 1e-05,
3895
+ "steps_completed": 26,
3896
+ "peak_reward": 1.0,
3897
+ "last10_avg": 0.9791666666666666,
3898
+ "first5_avg": 0.975,
3899
+ "reward_trace": [
3900
+ 0.875,
3901
+ 1.0,
3902
+ 1.0,
3903
+ 1.0,
3904
+ 1.0,
3905
+ 1.0
3906
+ ],
3907
+ "status": "completed",
3908
+ "source": "bitter_lesson_campaign_v2"
3909
+ },
3910
+ {
3911
+ "experiment_id": "campaign_v2_w2_qwen3-8b_G16",
3912
+ "model": "Qwen/Qwen3-8B",
3913
+ "method": "GRPO",
3914
+ "platform": "Tinker",
3915
+ "task": "gsm8k",
3916
+ "seed": 42,
3917
+ "group_size": 16,
3918
+ "lr": 1e-05,
3919
+ "steps_completed": 26,
3920
+ "peak_reward": 0.719,
3921
+ "last10_avg": 0.38033333333333336,
3922
+ "first5_avg": 0.4064,
3923
+ "reward_trace": [
3924
+ 0.719,
3925
+ 0.344,
3926
+ 0.25,
3927
+ 0.125,
3928
+ 0.594,
3929
+ 0.25
3930
+ ],
3931
+ "status": "completed",
3932
+ "source": "bitter_lesson_campaign_v2"
3933
+ },
3934
+ {
3935
+ "experiment_id": "campaign_v2_w2_qwen3-8b_G2",
3936
+ "model": "Qwen/Qwen3-8B",
3937
+ "method": "GRPO",
3938
+ "platform": "Tinker",
3939
+ "task": "gsm8k",
3940
+ "seed": 42,
3941
+ "group_size": 2,
3942
+ "lr": 1e-05,
3943
+ "steps_completed": 26,
3944
+ "peak_reward": 0.5,
3945
+ "last10_avg": 0.375,
3946
+ "first5_avg": 0.35,
3947
+ "reward_trace": [
3948
+ 0.0,
3949
+ 0.5,
3950
+ 0.5,
3951
+ 0.25,
3952
+ 0.5,
3953
+ 0.5
3954
+ ],
3955
+ "status": "completed",
3956
+ "source": "bitter_lesson_campaign_v2"
3957
+ },
3958
+ {
3959
+ "experiment_id": "campaign_v2_w2_qwen3-8b_G32",
3960
+ "model": "Qwen/Qwen3-8B",
3961
+ "method": "GRPO",
3962
+ "platform": "Tinker",
3963
+ "task": "gsm8k",
3964
+ "seed": 42,
3965
+ "group_size": 32,
3966
+ "lr": 1e-05,
3967
+ "steps_completed": 11,
3968
+ "peak_reward": 0.453,
3969
+ "last10_avg": 0.43766666666666665,
3970
+ "first5_avg": 0.43766666666666665,
3971
+ "reward_trace": [
3972
+ 0.438,
3973
+ 0.453,
3974
+ 0.422
3975
+ ],
3976
+ "status": "in_progress",
3977
+ "source": "bitter_lesson_campaign_v2"
3978
+ },
3979
+ {
3980
+ "experiment_id": "campaign_v2_w2_qwen3-8b_G4",
3981
+ "model": "Qwen/Qwen3-8B",
3982
+ "method": "GRPO",
3983
+ "platform": "Tinker",
3984
+ "task": "gsm8k",
3985
+ "seed": 42,
3986
+ "group_size": 4,
3987
+ "lr": 1e-05,
3988
+ "steps_completed": 26,
3989
+ "peak_reward": 0.75,
3990
+ "last10_avg": 0.5208333333333334,
3991
+ "first5_avg": 0.5,
3992
+ "reward_trace": [
3993
+ 0.75,
3994
+ 0.5,
3995
+ 0.25,
3996
+ 0.625,
3997
+ 0.375,
3998
+ 0.625
3999
+ ],
4000
+ "status": "completed",
4001
+ "source": "bitter_lesson_campaign_v2"
4002
+ },
4003
+ {
4004
+ "experiment_id": "campaign_v2_w3_dsv31_s123",
4005
+ "model": "deepseek-ai/DeepSeek-V3.1",
4006
+ "method": "GRPO",
4007
+ "platform": "Tinker",
4008
+ "task": "gsm8k",
4009
+ "seed": 123,
4010
+ "group_size": 8,
4011
+ "lr": 1e-05,
4012
+ "steps_completed": 1,
4013
+ "peak_reward": 0.875,
4014
+ "last10_avg": 0.875,
4015
+ "first5_avg": 0.875,
4016
+ "reward_trace": [
4017
+ 0.875
4018
+ ],
4019
+ "status": "in_progress",
4020
+ "source": "bitter_lesson_campaign_v2"
4021
+ },
4022
+ {
4023
+ "experiment_id": "campaign_v2_w3_dsv31_s456",
4024
+ "model": "deepseek-ai/DeepSeek-V3.1",
4025
+ "method": "GRPO",
4026
+ "platform": "Tinker",
4027
+ "task": "gsm8k",
4028
+ "seed": 456,
4029
+ "group_size": 8,
4030
+ "lr": 1e-05,
4031
+ "steps_completed": 1,
4032
+ "peak_reward": 1.0,
4033
+ "last10_avg": 1.0,
4034
+ "first5_avg": 1.0,
4035
+ "reward_trace": [
4036
+ 1.0
4037
+ ],
4038
+ "status": "in_progress",
4039
+ "source": "bitter_lesson_campaign_v2"
4040
+ },
4041
+ {
4042
+ "tag": "trl_qwen3_8b",
4043
+ "model": "Qwen/Qwen3-8B",
4044
+ "model_short": "qwen3-8b",
4045
+ "method": "TRL-GRPO",
4046
+ "task": "gsm8k",
4047
+ "platform": "modal_h100",
4048
+ "peak_reward": 0.375,
4049
+ "last10_avg": 0.05,
4050
+ "first5_avg": 0.0,
4051
+ "reward_trace": [
4052
+ 0.0,
4053
+ 0.0,
4054
+ 0.0,
4055
+ 0.0,
4056
+ 0.0,
4057
+ 0.125,
4058
+ 0.125,
4059
+ 0.0,
4060
+ 0.0,
4061
+ 0.0,
4062
+ 0.0,
4063
+ 0.0,
4064
+ 0.125,
4065
+ 0.0,
4066
+ 0.0,
4067
+ 0.0,
4068
+ 0.0,
4069
+ 0.0,
4070
+ 0.0,
4071
+ 0.0,
4072
+ 0.0,
4073
+ 0.0,
4074
+ 0.0,
4075
+ 0.375,
4076
+ 0.0,
4077
+ 0.125,
4078
+ 0.0,
4079
+ 0.0,
4080
+ 0.0,
4081
+ 0.0
4082
+ ],
4083
+ "duration_s": 735.7421374320984,
4084
+ "num_steps": 30,
4085
+ "status": "completed",
4086
+ "hf_model": "arvindcr4/trl_qwen3_8b-grpo-gsm8k"
4087
+ },
4088
+ {
4089
+ "tag": "trl_llama32_3b",
4090
+ "model": "meta-llama/Llama-3.2-3B-Instruct",
4091
+ "model_short": "llama-3.2-3b",
4092
+ "method": "TRL-GRPO",
4093
+ "task": "gsm8k",
4094
+ "platform": "modal_h100",
4095
+ "peak_reward": 1.0,
4096
+ "last10_avg": 0.7125,
4097
+ "first5_avg": 0.675,
4098
+ "reward_trace": [
4099
+ 0.625,
4100
+ 0.5,
4101
+ 0.625,
4102
+ 0.75,
4103
+ 0.875,
4104
+ 1.0,
4105
+ 1.0,
4106
+ 0.625,
4107
+ 1.0,
4108
+ 0.875,
4109
+ 0.5,
4110
+ 0.875,
4111
+ 0.75,
4112
+ 0.875,
4113
+ 0.375,
4114
+ 1.0,
4115
+ 0.875,
4116
+ 0.75,
4117
+ 1.0,
4118
+ 0.25,
4119
+ 0.75,
4120
+ 0.5,
4121
+ 0.75,
4122
+ 0.875,
4123
+ 0.125,
4124
+ 1.0,
4125
+ 0.625,
4126
+ 1.0,
4127
+ 0.625,
4128
+ 0.875
4129
+ ],
4130
+ "duration_s": 222.11492276191711,
4131
+ "num_steps": 30,
4132
+ "status": "completed",
4133
+ "hf_model": "arvindcr4/trl_llama32_3b-grpo-gsm8k"
4134
+ },
4135
+ {
4136
+ "tag": "trl_llama32_1b",
4137
+ "model": "meta-llama/Llama-3.2-1B-Instruct",
4138
+ "model_short": "llama-3.2-1b",
4139
+ "method": "TRL-GRPO",
4140
+ "task": "gsm8k",
4141
+ "platform": "modal_h100",
4142
+ "peak_reward": 0.875,
4143
+ "last10_avg": 0.3625,
4144
+ "first5_avg": 0.175,
4145
+ "reward_trace": [
4146
+ 0.125,
4147
+ 0.0,
4148
+ 0.0,
4149
+ 0.125,
4150
+ 0.625,
4151
+ 0.375,
4152
+ 0.125,
4153
+ 0.125,
4154
+ 0.0,
4155
+ 0.125,
4156
+ 0.125,
4157
+ 0.0,
4158
+ 0.0,
4159
+ 0.25,
4160
+ 0.0,
4161
+ 0.375,
4162
+ 0.125,
4163
+ 0.0,
4164
+ 0.375,
4165
+ 0.0,
4166
+ 0.0,
4167
+ 0.625,
4168
+ 0.625,
4169
+ 0.125,
4170
+ 0.125,
4171
+ 0.375,
4172
+ 0.5,
4173
+ 0.875,
4174
+ 0.25,
4175
+ 0.125
4176
+ ],
4177
+ "duration_s": 162.25860285758972,
4178
+ "num_steps": 30,
4179
+ "status": "completed",
4180
+ "hf_model": "arvindcr4/trl_llama32_1b-grpo-gsm8k"
4181
+ }
4182
+ ]
4183
+ }
training_runs.jsonl ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"training_run_id": "c465d654-a00f-5821-9c11-0e40f0eecf9f", "experiment": "arch_gsm8k_gpt-oss-20b", "model": "openai/gpt-oss-20b", "model_short": "gpt-oss-20b", "task": "gsm8k", "seed": 42, "rank": 32, "lr": 3e-05, "group_size": 8, "steps": 30, "platform": null, "last10_avg": null, "peak_accuracy": null, "last10_accuracy": null, "timestamp": null, "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://c465d654-a00f-5821-9c11-0e40f0eecf9f:train:0/sampler_weights/final"}]}
2
+ {"training_run_id": "51a8ef9e-15ef-5f8f-bda1-78ee51387a12", "experiment": "arch_gsm8k_kimi-k2", "model": "moonshotai/Kimi-K2-Thinking", "model_short": "kimi-k2", "task": "gsm8k", "seed": 42, "rank": 16, "lr": 1e-05, "group_size": 4, "steps": 20, "platform": null, "last10_avg": null, "peak_accuracy": null, "last10_accuracy": null, "timestamp": null, "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://51a8ef9e-15ef-5f8f-bda1-78ee51387a12:train:0/sampler_weights/final"}]}
3
+ {"training_run_id": "ca2e3a24-7401-5770-af34-a0d27177aeaa", "experiment": "cross_tool_llama-8b-inst", "model": "meta-llama/Llama-3.1-8B-Instruct", "model_short": "llama-8b-inst", "task": "tool_use", "seed": 42, "rank": 32, "lr": 3e-05, "group_size": 8, "steps": 30, "platform": "tinker", "last10_avg": 0.0, "peak_accuracy": 0.0, "last10_accuracy": 0.0, "timestamp": "2026-04-18T16:01:52.013977", "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://ca2e3a24-7401-5770-af34-a0d27177aeaa:train:0/sampler_weights/final"}]}
4
+ {"training_run_id": "73ff186a-d8e3-50c3-afb8-2b863cd09579", "experiment": "cross_tool_qwen3-32b", "model": "Qwen/Qwen3-32B", "model_short": "qwen3-32b", "task": "tool_use", "seed": 42, "rank": 32, "lr": 3e-05, "group_size": 8, "steps": 30, "platform": "tinker", "last10_avg": 0.0, "peak_accuracy": 0.0, "last10_accuracy": 0.0, "timestamp": "2026-04-19T03:41:34.408077", "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://73ff186a-d8e3-50c3-afb8-2b863cd09579:train:0/sampler_weights/final"}]}
5
+ {"training_run_id": "56b99b24-03bd-5d00-ae23-831933ef53b2", "experiment": "frontier_gsm8k_deepseek-v3.1", "model": "deepseek-ai/DeepSeek-V3.1", "model_short": "deepseek-v3.1", "task": "gsm8k", "seed": 42, "rank": 16, "lr": 1e-05, "group_size": 4, "steps": 20, "platform": "tinker", "last10_avg": 0.85, "peak_accuracy": 1.0, "last10_accuracy": 0.85, "timestamp": "2026-04-18T16:05:46.209173", "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://56b99b24-03bd-5d00-ae23-831933ef53b2:train:0/sampler_weights/final"}]}
6
+ {"training_run_id": "657a920a-9e74-55d2-9354-71a6ec2f1f61", "experiment": "frontier_gsm8k_nemotron-120b", "model": "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16", "model_short": "nemotron-120b", "task": "gsm8k", "seed": 42, "rank": 16, "lr": 1e-05, "group_size": 4, "steps": 20, "platform": "tinker", "last10_avg": 0.1625, "peak_accuracy": 0.875, "last10_accuracy": 0.1625, "timestamp": "2026-04-19T03:31:55.848278", "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://657a920a-9e74-55d2-9354-71a6ec2f1f61:train:0/sampler_weights/final"}]}
7
+ {"training_run_id": "488bbb2e-aa35-5431-b4a5-793f68b6bc49", "experiment": "scale_gsm8k_llama-8b-inst", "model": "meta-llama/Llama-3.1-8B-Instruct", "model_short": "llama-8b-inst", "task": "gsm8k", "seed": 42, "rank": 32, "lr": 3e-05, "group_size": 8, "steps": 30, "platform": "tinker", "last10_avg": 0.84375, "peak_accuracy": 1.0, "last10_accuracy": 0.84375, "timestamp": "2026-04-19T04:11:44.481966", "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://488bbb2e-aa35-5431-b4a5-793f68b6bc49:train:0/sampler_weights/final"}]}
8
+ {"training_run_id": "3154dee5-4fd7-59b6-ba3b-721eef675bfc", "experiment": "scale_gsm8k_qwen3-8b", "model": "Qwen/Qwen3-8B", "model_short": "qwen3-8b", "task": "gsm8k", "seed": 42, "rank": 32, "lr": 3e-05, "group_size": 8, "steps": 30, "platform": "tinker", "last10_avg": 0.34375, "peak_accuracy": 0.625, "last10_accuracy": 0.34375, "timestamp": "2026-04-18T16:31:17.561535", "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://3154dee5-4fd7-59b6-ba3b-721eef675bfc:train:0/sampler_weights/final"}]}
9
+ {"training_run_id": "200e4983-3e9d-5176-87d0-c55dd3c73539", "experiment": "scale_gsm8k_qwen3.5-4b", "model": "Qwen/Qwen3.5-4B", "model_short": "qwen3.5-4b", "task": "gsm8k", "seed": 42, "rank": 32, "lr": 3e-05, "group_size": 8, "steps": 30, "platform": "tinker", "last10_avg": 0.85, "peak_accuracy": 1.0, "last10_accuracy": 0.85, "timestamp": "2026-04-19T03:34:22.495977", "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://200e4983-3e9d-5176-87d0-c55dd3c73539:train:0/sampler_weights/final"}]}
10
+ {"training_run_id": "0416d5d1-161b-5264-8b69-4612314d3b41", "experiment": null, "model": "deepseek-ai/DeepSeek-V3.1", "model_short": null, "task": null, "seed": 456, "rank": null, "lr": 1e-05, "group_size": null, "steps": 30, "platform": null, "last10_avg": null, "peak_accuracy": null, "last10_accuracy": null, "timestamp": null, "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://0416d5d1-161b-5264-8b69-4612314d3b41:train:0/sampler_weights/final"}]}
11
+ {"training_run_id": "16d810a0-9818-5f4c-820c-23fc91270769", "experiment": null, "model": "Qwen/Qwen3-8B", "model_short": null, "task": null, "seed": 42, "rank": null, "lr": 1e-05, "group_size": null, "steps": 30, "platform": null, "last10_avg": null, "peak_accuracy": null, "last10_accuracy": null, "timestamp": null, "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://16d810a0-9818-5f4c-820c-23fc91270769:train:0/sampler_weights/final"}]}
12
+ {"training_run_id": "200252fc-0649-5b39-8030-ba0fa5d07e84", "experiment": null, "model": "deepseek-ai/DeepSeek-V3.1-Base", "model_short": null, "task": null, "seed": 42, "rank": null, "lr": 1e-05, "group_size": null, "steps": 30, "platform": null, "last10_avg": null, "peak_accuracy": null, "last10_accuracy": null, "timestamp": null, "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://200252fc-0649-5b39-8030-ba0fa5d07e84:train:0/sampler_weights/final"}]}
13
+ {"training_run_id": "380ee7fe-a0fc-5224-b755-49256a020831", "experiment": null, "model": null, "model_short": null, "task": null, "seed": null, "rank": null, "lr": null, "group_size": null, "steps": null, "platform": null, "last10_avg": null, "peak_accuracy": null, "last10_accuracy": null, "timestamp": null, "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "step_50", "uri": "tinker://380ee7fe-a0fc-5224-b755-49256a020831:train:0/sampler_weights/step_50"}]}
14
+ {"training_run_id": "38d13280-7597-5485-ac06-0427525fc7ae", "experiment": null, "model": null, "model_short": null, "task": null, "seed": null, "rank": null, "lr": null, "group_size": null, "steps": null, "platform": null, "last10_avg": null, "peak_accuracy": null, "last10_accuracy": null, "timestamp": null, "weight_checkpoints": [{"step": 100, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000100"}, {"step": 150, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000150"}, {"step": 200, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000200"}, {"step": 250, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000250"}, {"step": 300, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000300"}, {"step": 350, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000350"}, {"step": 400, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000400"}, {"step": 450, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000450"}, {"step": 50, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000050"}, {"step": 500, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000500"}, {"step": 550, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000550"}, {"step": 600, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000600"}, {"step": 650, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000650"}, {"step": 700, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/weights/000700"}], "sampler_weight_checkpoints": [{"step": 100, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000100"}, {"step": 150, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000150"}, {"step": 200, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000200"}, {"step": 250, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000250"}, {"step": 300, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000300"}, {"step": 350, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000350"}, {"step": 400, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000400"}, {"step": 450, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000450"}, {"step": 50, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000050"}, {"step": 500, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000500"}, {"step": 550, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000550"}, {"step": 600, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000600"}, {"step": 650, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000650"}, {"step": 700, "uri": "tinker://38d13280-7597-5485-ac06-0427525fc7ae:train:0/sampler_weights/000700"}]}
15
+ {"training_run_id": "39aa5eb2-e234-5a95-ab68-896e4cac8c45", "experiment": null, "model": null, "model_short": null, "task": null, "seed": null, "rank": null, "lr": null, "group_size": null, "steps": null, "platform": null, "last10_avg": null, "peak_accuracy": null, "last10_accuracy": null, "timestamp": null, "weight_checkpoints": [{"step": 100, "uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/weights/000100"}, {"step": 20, "uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/weights/000020"}, {"step": 40, "uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/weights/000040"}, {"step": 60, "uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/weights/000060"}, {"step": 80, "uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/weights/000080"}, {"step": "final", "uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/weights/final"}], "sampler_weight_checkpoints": [{"step": 100, "uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/sampler_weights/000100"}, {"step": 20, "uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/sampler_weights/000020"}, {"step": 40, "uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/sampler_weights/000040"}, {"step": 60, "uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/sampler_weights/000060"}, {"step": 80, "uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/sampler_weights/000080"}, {"step": "final", "uri": "tinker://39aa5eb2-e234-5a95-ab68-896e4cac8c45:train:0/sampler_weights/final"}]}
16
+ {"training_run_id": "3fd600fd-6982-5061-8edc-c2da2c7f2f0e", "experiment": null, "model": "deepseek-ai/DeepSeek-V3.1", "model_short": null, "task": null, "seed": 789, "rank": null, "lr": 1e-05, "group_size": null, "steps": 30, "platform": null, "last10_avg": null, "peak_accuracy": null, "last10_accuracy": null, "timestamp": null, "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://3fd600fd-6982-5061-8edc-c2da2c7f2f0e:train:0/sampler_weights/final"}]}
17
+ {"training_run_id": "5761c236-774f-5b29-926e-13bd6af310b1", "experiment": null, "model": "Qwen/Qwen3-8B", "model_short": null, "task": null, "seed": 42, "rank": null, "lr": 1e-05, "group_size": null, "steps": 30, "platform": null, "last10_avg": null, "peak_accuracy": null, "last10_accuracy": null, "timestamp": null, "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://5761c236-774f-5b29-926e-13bd6af310b1:train:0/sampler_weights/final"}]}
18
+ {"training_run_id": "57f4d068-2997-577c-895e-05f2265935a5", "experiment": null, "model": "meta-llama/Llama-3.1-8B", "model_short": null, "task": null, "seed": 42, "rank": null, "lr": 1e-05, "group_size": null, "steps": 30, "platform": null, "last10_avg": null, "peak_accuracy": null, "last10_accuracy": null, "timestamp": null, "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://57f4d068-2997-577c-895e-05f2265935a5:train:0/sampler_weights/final"}]}
19
+ {"training_run_id": "7040f210-e755-5f8d-bd1a-5cb6f3ab33f8", "experiment": null, "model": "Qwen/Qwen3.5-397B-A17B", "model_short": null, "task": null, "seed": 42, "rank": null, "lr": 1e-05, "group_size": null, "steps": 30, "platform": null, "last10_avg": null, "peak_accuracy": null, "last10_accuracy": null, "timestamp": null, "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://7040f210-e755-5f8d-bd1a-5cb6f3ab33f8:train:0/sampler_weights/final"}]}
20
+ {"training_run_id": "9f27c001-b92c-55a7-9e12-1a8bd858e16d", "experiment": null, "model": "Qwen/Qwen3-8B-Base", "model_short": null, "task": null, "seed": 42, "rank": null, "lr": 1e-05, "group_size": null, "steps": 30, "platform": null, "last10_avg": null, "peak_accuracy": null, "last10_accuracy": null, "timestamp": null, "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://9f27c001-b92c-55a7-9e12-1a8bd858e16d:train:0/sampler_weights/final"}]}
21
+ {"training_run_id": "a0d96a00-54e1-569c-a56c-98520819f14f", "experiment": null, "model": "openai/gpt-oss-120b", "model_short": null, "task": null, "seed": 42, "rank": null, "lr": 1e-05, "group_size": null, "steps": 30, "platform": null, "last10_avg": null, "peak_accuracy": null, "last10_accuracy": null, "timestamp": null, "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://a0d96a00-54e1-569c-a56c-98520819f14f:train:0/sampler_weights/final"}]}
22
+ {"training_run_id": "a1d398a3-cea7-55fd-97d9-8b61001550b2", "experiment": null, "model": "moonshotai/Kimi-K2.5", "model_short": null, "task": null, "seed": 42, "rank": null, "lr": 1e-05, "group_size": null, "steps": 30, "platform": null, "last10_avg": null, "peak_accuracy": null, "last10_accuracy": null, "timestamp": null, "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://a1d398a3-cea7-55fd-97d9-8b61001550b2:train:0/sampler_weights/final"}]}
23
+ {"training_run_id": "a5715c67-65ef-5db8-a5ce-6d115a04b874", "experiment": null, "model": "moonshotai/Kimi-K2-Thinking", "model_short": null, "task": null, "seed": 42, "rank": null, "lr": 1e-05, "group_size": null, "steps": 30, "platform": null, "last10_avg": null, "peak_accuracy": null, "last10_accuracy": null, "timestamp": null, "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://a5715c67-65ef-5db8-a5ce-6d115a04b874:train:0/sampler_weights/final"}]}
24
+ {"training_run_id": "b39ea661-c094-5ecb-bed7-925c3e22f285", "experiment": null, "model": "meta-llama/Llama-3.1-70B", "model_short": null, "task": null, "seed": 42, "rank": null, "lr": 1e-05, "group_size": null, "steps": 30, "platform": null, "last10_avg": null, "peak_accuracy": null, "last10_accuracy": null, "timestamp": null, "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://b39ea661-c094-5ecb-bed7-925c3e22f285:train:0/sampler_weights/final"}]}
25
+ {"training_run_id": "c589e374-95d8-520d-8a4e-1b688bc5550f", "experiment": null, "model": "deepseek-ai/DeepSeek-V3.1", "model_short": null, "task": null, "seed": 123, "rank": null, "lr": 1e-05, "group_size": null, "steps": 30, "platform": null, "last10_avg": null, "peak_accuracy": null, "last10_accuracy": null, "timestamp": null, "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://c589e374-95d8-520d-8a4e-1b688bc5550f:train:0/sampler_weights/final"}]}
26
+ {"training_run_id": "c82c876f-b1ff-517c-b19b-472b3df8e4b4", "experiment": null, "model": "Qwen/Qwen3-8B", "model_short": null, "task": null, "seed": 42, "rank": null, "lr": 1e-05, "group_size": null, "steps": 30, "platform": null, "last10_avg": null, "peak_accuracy": null, "last10_accuracy": null, "timestamp": null, "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://c82c876f-b1ff-517c-b19b-472b3df8e4b4:train:0/sampler_weights/final"}]}
27
+ {"training_run_id": "ca0d6a00-3a4c-5a24-8b8d-f9d69db9609e", "experiment": null, "model": "Qwen/Qwen3-8B-Base", "model_short": null, "task": null, "seed": 42, "rank": null, "lr": 1e-05, "group_size": null, "steps": 30, "platform": null, "last10_avg": null, "peak_accuracy": null, "last10_accuracy": null, "timestamp": null, "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://ca0d6a00-3a4c-5a24-8b8d-f9d69db9609e:train:0/sampler_weights/final"}]}
28
+ {"training_run_id": "de39d965-b3f3-5317-b4e7-259a9472f2e0", "experiment": null, "model": "Qwen/Qwen3-8B", "model_short": null, "task": null, "seed": 42, "rank": null, "lr": 1e-05, "group_size": null, "steps": 30, "platform": null, "last10_avg": null, "peak_accuracy": null, "last10_accuracy": null, "timestamp": null, "weight_checkpoints": [], "sampler_weight_checkpoints": [{"step": "final", "uri": "tinker://de39d965-b3f3-5317-b4e7-259a9472f2e0:train:0/sampler_weights/final"}]}