Hugging Face's logo

tumorailab
/
IDPRO-ESMC600M

idpro
protein
language-model
esm
qwen
Model card Files
xet
 Community
sposhiy commited on
Commit
cd94445
·
verified ·
1 Parent(s): 9d3fc9a

Initial IDPro ESM C 600M Stage-4 step-80000 release

Browse files
Files changed (4) hide show
  1. README.md +88 -0
  2. config.json +66 -0
  3. idpro_state.pt +3 -0
  4. metadata.json +1246 -0
README.md ADDED
@@ -0,0 +1,88 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: idpro
3
+ license: cc-by-nc-4.0
4
+ tags:
5
+ - protein
6
+ - language-model
7
+ - idpro
8
+ - esm
9
+ - qwen
10
+ base_model:
11
+ - EvolutionaryScale/esmc-600m-2024-12
12
+ - Qwen/Qwen3.5-27B
13
+ ---
14
+
15
+ # IDPro — user/idpro-esmc-stage4
16
+
17
+ IDPro (Illuminating the Dark PROteome) is a multimodal protein → text model.
18
+ A frozen ESM C protein encoder feeds per-residue embeddings through a 1D-Conv
19
+ **adaptor** and an MLP **projector** into a Qwen3.5-27B language model. This
20
+ release is the **fully-trained Stage-4 ESM C arm** — the canonical baseline
21
+ used in the IDPro Aim 1 analyses (DOE Genesis Phase I).
22
+
23
+ ## What's in this release
24
+
25
+ The release ships only the **trained weights on top of the two frozen base
26
+ models** — the bases (ESM C 600M and Qwen3.5-27B) are not bundled.
27
+
28
+ | Component | Shape / size | Source |
29
+ |---|---|---|
30
+ | `adaptor` (2× 1D Conv, kernel 7) | encoder_dim=1152 | this release |
31
+ | `projector` (2-layer MLP) | 1152 → 5120 | this release |
32
+ | `protein_modality_embed` + `prot_end_embed` | (1, 1, 5120) each | this release |
33
+ | `protein_position` (sinusoidal + learned scale) | small | this release |
34
+ | `evidence_head_pre` (pre-LLM, on adaptor output) | 9-class | this release |
35
+ | `evidence_head_post` (post-LLM, layer 48 of Qwen) | 9-class | this release |
36
+ | `llm.lm_head.weight` (resized output projection) | (vocab × 5120) | this release |
37
+ | ESM C 600M encoder (frozen) | — | `EvolutionaryScale/esmc-600m-2024-12` |
38
+ | Qwen3.5-27B base LLM (frozen) | — | `Qwen/Qwen3.5-27B` |
39
+
40
+ `idpro_state.pt` is a `{name: tensor}` dict with ~29 keys. It is loaded with
41
+ `strict=False` against the IDPro module tree (the encoder and base LLM keys
42
+ are not present in this dict — they come from the frozen bases).
43
+
44
+ ## Usage
45
+
46
+ ```python
47
+ import torch
48
+ from idpro.model import IDProModel
49
+
50
+ model = IDProModel.from_pretrained(
51
+ "user/idpro-esmc-stage4",
52
+ device="cuda",
53
+ dtype=torch.bfloat16,
54
+ # llm_name="<local-qwen-path>", # optional: point at a local Qwen snapshot
55
+ )
56
+
57
+ answers = model.generate(
58
+ sequences=["MGSSHHHHHHSSGLVPRGSHM..."],
59
+ questions=["What is the function of this protein?"],
60
+ )
61
+ ```
62
+
63
+ The Aim 1 evaluation pipeline (probe benchmarks + conformal calibration) takes
64
+ a local checkpoint directory via `--ckpt`, so you can either point it at the
65
+ release dir directly or stage one with `huggingface_hub.snapshot_download`.
66
+
67
+ ## Special tokens
68
+
69
+ The training added four IDPro special tokens to the Qwen tokenizer:
70
+
71
+ `<PROT_START>`, `<PROT_END>`, `<RAG_START>`, `<RAG_END>`
72
+
73
+ `from_pretrained` rebuilds the tokenizer by adding these via
74
+ `tokenizer.add_special_tokens(...)` and resizing the model's token embeddings
75
+ to match the saved `lm_head` row count.
76
+
77
+ ## Provenance
78
+
79
+ - Source repo: doe_genesis (DOE Genesis Phase I, Focus Area 2C)
80
+ - Source checkpoint: `/data/asahu/projects/doe_genesis/idpro/checkpoints/robust/stage4_step80000`
81
+ - Stage / step: `stage4_step80000`
82
+ - Notes: Canonical IDPro release: ESM C 600M + Qwen3.5-27B, Stage 4 step 80,000. EC-L1 macro-AUC 0.920 on the IDPro reference benchmark.
83
+ - Eval (selected): 5-fold CV macro-AUC 0.944 (A_linear) / 0.956 (A+B+C_linear) on the EC-L1 reference benchmark.
84
+
85
+ ## License
86
+
87
+ Weights are released under CC-BY-NC-4.0 (research use; check upstream model
88
+ licenses for ESM C and Qwen3.5-27B before redistribution).
config.json ADDED
@@ -0,0 +1,66 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "encoder": {
3
+ "name": "EvolutionaryScale/esmc-600m-2024-12",
4
+ "dim": 1152,
5
+ "freeze": true,
6
+ "select_layer": -1,
7
+ "backend": "esmc",
8
+ "structure_track": false,
9
+ "structure_manifest_path": ""
10
+ },
11
+ "adaptor": {
12
+ "num_layers": 2,
13
+ "kernel_size": 7,
14
+ "dropout": 0.0
15
+ },
16
+ "projector": {
17
+ "num_layers": 2,
18
+ "activation": "gelu"
19
+ },
20
+ "llm": {
21
+ "name": "Qwen/Qwen3.5-27B",
22
+ "dim": 5120,
23
+ "max_context": 262144,
24
+ "lora_r": 64,
25
+ "lora_alpha": 128,
26
+ "lora_dropout": 0.05,
27
+ "lora_target_modules": [
28
+ "q_proj",
29
+ "k_proj",
30
+ "v_proj",
31
+ "o_proj",
32
+ "gate_proj",
33
+ "up_proj",
34
+ "down_proj"
35
+ ],
36
+ "dtype": "bf16",
37
+ "use_qlora": false
38
+ },
39
+ "rag": {
40
+ "enabled": true,
41
+ "k": 5,
42
+ "embedding_index_path": "",
43
+ "max_context_tokens": 512
44
+ },
45
+ "training": {
46
+ "stage": 1,
47
+ "train_adaptor": true,
48
+ "train_projector": true,
49
+ "train_llm_lora": false,
50
+ "freeze_encoder": true,
51
+ "lr": 0.002,
52
+ "batch_size": 4,
53
+ "grad_accum_steps": 8,
54
+ "epochs": 1,
55
+ "max_seq_len": 2048,
56
+ "warmup_ratio": 0.03,
57
+ "weight_decay": 0.01,
58
+ "lr_scheduler": "cosine",
59
+ "replay_fraction": 0.0,
60
+ "rag_fraction": 0.0
61
+ },
62
+ "prot_start_token": "<PROT_START>",
63
+ "prot_end_token": "<PROT_END>",
64
+ "rag_start_token": "<RAG_START>",
65
+ "rag_end_token": "<RAG_END>"
66
+ }
idpro_state.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4eb5269b7a246222fe04e9bab1085c6d85a2acf040f2bc9c63f11470e603dbad
3
+ size 2648276306
metadata.json ADDED
@@ -0,0 +1,1246 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "num_keys": 29,
3
+ "total_params": 1324133395,
4
+ "lm_head_vocab_size": 248081,
5
+ "key_groups": {
6
+ "adaptor": 8,
7
+ "projector": 4,
8
+ "evidence_head_pre": 6,
9
+ "evidence_head_post": 6,
10
+ "protein_position": 2,
11
+ "special_embeds": 2,
12
+ "lm_head": 1,
13
+ "lora": 0
14
+ },
15
+ "source_state": {
16
+ "stage": 4,
17
+ "step": 80000,
18
+ "eval_log": [
19
+ {
20
+ "stage": 4,
21
+ "step": 0,
22
+ "f1": 0.0,
23
+ "time_min": 0,
24
+ "loss": 0
25
+ },
26
+ {
27
+ "stage": 4,
28
+ "step": 500,
29
+ "f1": 0.0,
30
+ "time_min": 6.855769292513529,
31
+ "loss": 1.1732854264974595
32
+ },
33
+ {
34
+ "stage": 4,
35
+ "step": 1000,
36
+ "f1": 0.0014492753623188408,
37
+ "time_min": 23.952689743041994,
38
+ "loss": 1.0679561794400214
39
+ },
40
+ {
41
+ "stage": 4,
42
+ "step": 1500,
43
+ "f1": 0.0,
44
+ "time_min": 40.45468080838521,
45
+ "loss": 1.0036508800387383
46
+ },
47
+ {
48
+ "stage": 4,
49
+ "step": 2000,
50
+ "f1": 0.005066538399871733,
51
+ "time_min": 57.2181734363238,
52
+ "loss": 0.9658531248271466
53
+ },
54
+ {
55
+ "stage": 4,
56
+ "step": 2000,
57
+ "f1": 0.0011299435028248588,
58
+ "time_min": 0,
59
+ "loss": 0
60
+ },
61
+ {
62
+ "stage": 4,
63
+ "step": 2500,
64
+ "f1": 0.007141092167037312,
65
+ "time_min": 6.157433672746023,
66
+ "loss": 1.093848851084709
67
+ },
68
+ {
69
+ "stage": 4,
70
+ "step": 3000,
71
+ "f1": 0.0,
72
+ "time_min": 20.813616796334586,
73
+ "loss": 1.06912658944726
74
+ },
75
+ {
76
+ "stage": 4,
77
+ "step": 3500,
78
+ "f1": 0.0,
79
+ "time_min": 37.01884375810623,
80
+ "loss": 1.0461089491645494
81
+ },
82
+ {
83
+ "stage": 4,
84
+ "step": 4000,
85
+ "f1": 0.007897435897435898,
86
+ "time_min": 51.83560905853907,
87
+ "loss": 1.0333229052126407
88
+ },
89
+ {
90
+ "stage": 4,
91
+ "step": 4500,
92
+ "f1": 0.008333969587071326,
93
+ "time_min": 66.55019008715948,
94
+ "loss": 1.0255064507722855
95
+ },
96
+ {
97
+ "stage": 4,
98
+ "step": 5000,
99
+ "f1": 0.007648953301127214,
100
+ "time_min": 81.21829265753428,
101
+ "loss": 1.0210118336280187
102
+ },
103
+ {
104
+ "stage": 4,
105
+ "step": 5500,
106
+ "f1": 0.00299625468164794,
107
+ "time_min": 95.43870557546616,
108
+ "loss": 1.01732074204513
109
+ },
110
+ {
111
+ "stage": 4,
112
+ "step": 6000,
113
+ "f1": 0.0,
114
+ "time_min": 110.34333517948787,
115
+ "loss": 1.0131796339303256
116
+ },
117
+ {
118
+ "stage": 4,
119
+ "step": 6500,
120
+ "f1": 0.002380952380952381,
121
+ "time_min": 125.31258865992228,
122
+ "loss": 1.0096191558440526
123
+ },
124
+ {
125
+ "stage": 4,
126
+ "step": 7000,
127
+ "f1": 0.007161803713527852,
128
+ "time_min": 139.8337462147077,
129
+ "loss": 1.004719085508585
130
+ },
131
+ {
132
+ "stage": 4,
133
+ "step": 7500,
134
+ "f1": 0.0,
135
+ "time_min": 154.31689653396606,
136
+ "loss": 1.0029836373871024
137
+ },
138
+ {
139
+ "stage": 4,
140
+ "step": 8000,
141
+ "f1": 0.0,
142
+ "time_min": 169.36405350764593,
143
+ "loss": 0.9982387975007295
144
+ },
145
+ {
146
+ "stage": 4,
147
+ "step": 8500,
148
+ "f1": 0.0015503875968992248,
149
+ "time_min": 184.0973814845085,
150
+ "loss": 0.9950762509703636
151
+ },
152
+ {
153
+ "stage": 4,
154
+ "step": 9000,
155
+ "f1": 0.0,
156
+ "time_min": 199.2812495748202,
157
+ "loss": 0.9945344764632839
158
+ },
159
+ {
160
+ "stage": 4,
161
+ "step": 9500,
162
+ "f1": 0.0,
163
+ "time_min": 213.7300217072169,
164
+ "loss": 0.9926016963362694
165
+ },
166
+ {
167
+ "stage": 4,
168
+ "step": 10000,
169
+ "f1": 0.0,
170
+ "time_min": 228.04795783758163,
171
+ "loss": 0.990872772295028
172
+ },
173
+ {
174
+ "stage": 4,
175
+ "step": 10500,
176
+ "f1": 0.0,
177
+ "time_min": 242.9892031232516,
178
+ "loss": 0.9901926384883768
179
+ },
180
+ {
181
+ "stage": 4,
182
+ "step": 11000,
183
+ "f1": 0.0,
184
+ "time_min": 257.94673357407254,
185
+ "loss": 0.98806184556418
186
+ },
187
+ {
188
+ "stage": 4,
189
+ "step": 11500,
190
+ "f1": 0.0,
191
+ "time_min": 272.2741646329562,
192
+ "loss": 0.9864317376174425
193
+ },
194
+ {
195
+ "stage": 4,
196
+ "step": 12000,
197
+ "f1": 0.0020512820512820513,
198
+ "time_min": 287.213184261322,
199
+ "loss": 0.984835107344389
200
+ },
201
+ {
202
+ "stage": 4,
203
+ "step": 12500,
204
+ "f1": 0.006481481481481482,
205
+ "time_min": 301.9303883989652,
206
+ "loss": 0.9836810060484069
207
+ },
208
+ {
209
+ "stage": 4,
210
+ "step": 13000,
211
+ "f1": 0.0,
212
+ "time_min": 316.10822058518727,
213
+ "loss": 0.9824496570094066
214
+ },
215
+ {
216
+ "stage": 4,
217
+ "step": 13500,
218
+ "f1": 0.0027598566308243725,
219
+ "time_min": 330.94340486129124,
220
+ "loss": 0.9811148852006248
221
+ },
222
+ {
223
+ "stage": 4,
224
+ "step": 14000,
225
+ "f1": 0.00455005055611729,
226
+ "time_min": 345.94701433579127,
227
+ "loss": 0.980539373010397
228
+ },
229
+ {
230
+ "stage": 4,
231
+ "step": 14500,
232
+ "f1": 0.0,
233
+ "time_min": 361.0001788338025,
234
+ "loss": 0.9795160161709785
235
+ },
236
+ {
237
+ "stage": 4,
238
+ "step": 15000,
239
+ "f1": 0.004537612146307798,
240
+ "time_min": 375.92283029556273,
241
+ "loss": 0.9772609019073156
242
+ },
243
+ {
244
+ "stage": 4,
245
+ "step": 15500,
246
+ "f1": 0.0017094017094017094,
247
+ "time_min": 390.77964694102604,
248
+ "loss": 0.9765125209159321
249
+ },
250
+ {
251
+ "stage": 4,
252
+ "step": 16000,
253
+ "f1": 0.0014336917562724012,
254
+ "time_min": 405.44099973837535,
255
+ "loss": 0.975278540992311
256
+ },
257
+ {
258
+ "stage": 4,
259
+ "step": 16500,
260
+ "f1": 0.007545605306799336,
261
+ "time_min": 420.0099479357402,
262
+ "loss": 0.9741899696136344
263
+ },
264
+ {
265
+ "stage": 4,
266
+ "step": 17000,
267
+ "f1": 0.0017852948085506225,
268
+ "time_min": 434.82097015778226,
269
+ "loss": 0.9726853859742483
270
+ },
271
+ {
272
+ "stage": 4,
273
+ "step": 17500,
274
+ "f1": 0.0008032128514056224,
275
+ "time_min": 449.8655348777771,
276
+ "loss": 0.9712435904914333
277
+ },
278
+ {
279
+ "stage": 4,
280
+ "step": 18000,
281
+ "f1": 0.0013333333333333333,
282
+ "time_min": 464.0333917697271,
283
+ "loss": 0.9699472135975957
284
+ },
285
+ {
286
+ "stage": 4,
287
+ "step": 18500,
288
+ "f1": 0.0,
289
+ "time_min": 478.93024856646855,
290
+ "loss": 0.9696770128571626
291
+ },
292
+ {
293
+ "stage": 4,
294
+ "step": 19000,
295
+ "f1": 0.0016666666666666666,
296
+ "time_min": 493.72316179275515,
297
+ "loss": 0.9689625195878393
298
+ },
299
+ {
300
+ "stage": 4,
301
+ "step": 19500,
302
+ "f1": 0.0,
303
+ "time_min": 508.4606335639954,
304
+ "loss": 0.9682910708001682
305
+ },
306
+ {
307
+ "stage": 4,
308
+ "step": 20000,
309
+ "f1": 0.0018018018018018018,
310
+ "time_min": 522.9376392960548,
311
+ "loss": 0.9679154083381096
312
+ },
313
+ {
314
+ "stage": 4,
315
+ "step": 20500,
316
+ "f1": 0.0004761904761904762,
317
+ "time_min": 537.8397751490276,
318
+ "loss": 0.967118464410305
319
+ },
320
+ {
321
+ "stage": 4,
322
+ "step": 21000,
323
+ "f1": 0.0,
324
+ "time_min": 552.5966074426968,
325
+ "loss": 0.9666775196288762
326
+ },
327
+ {
328
+ "stage": 4,
329
+ "step": 21500,
330
+ "f1": 0.0033333333333333335,
331
+ "time_min": 567.1978101650874,
332
+ "loss": 0.9659652575575388
333
+ },
334
+ {
335
+ "stage": 4,
336
+ "step": 22000,
337
+ "f1": 0.0,
338
+ "time_min": 582.1328752001126,
339
+ "loss": 0.9652409713700414
340
+ },
341
+ {
342
+ "stage": 4,
343
+ "step": 22500,
344
+ "f1": 0.000966183574879227,
345
+ "time_min": 596.5833401958148,
346
+ "loss": 0.9649073508236466
347
+ },
348
+ {
349
+ "stage": 4,
350
+ "step": 23000,
351
+ "f1": 0.0,
352
+ "time_min": 611.4143996596337,
353
+ "loss": 0.9649063978237765
354
+ },
355
+ {
356
+ "stage": 4,
357
+ "step": 23500,
358
+ "f1": 0.005152979066022545,
359
+ "time_min": 626.1295392553012,
360
+ "loss": 0.9643519908272943
361
+ },
362
+ {
363
+ "stage": 4,
364
+ "step": 24000,
365
+ "f1": 0.0,
366
+ "time_min": 640.8495065490405,
367
+ "loss": 0.9638447769826108
368
+ },
369
+ {
370
+ "stage": 4,
371
+ "step": 24500,
372
+ "f1": 0.0011299435028248588,
373
+ "time_min": 655.7232419053713,
374
+ "loss": 0.9632522215207417
375
+ },
376
+ {
377
+ "stage": 4,
378
+ "step": 25000,
379
+ "f1": 0.005530187721968544,
380
+ "time_min": 670.5895378271739,
381
+ "loss": 0.9622944644080555
382
+ },
383
+ {
384
+ "stage": 4,
385
+ "step": 25500,
386
+ "f1": 0.0,
387
+ "time_min": 685.3038972099622,
388
+ "loss": 0.9620724564676589
389
+ },
390
+ {
391
+ "stage": 4,
392
+ "step": 26000,
393
+ "f1": 0.001967081493376154,
394
+ "time_min": 700.1100624759991,
395
+ "loss": 0.9615654757022858
396
+ },
397
+ {
398
+ "stage": 4,
399
+ "step": 26500,
400
+ "f1": 0.0037692823608316568,
401
+ "time_min": 714.8974169572194,
402
+ "loss": 0.9612755754723841
403
+ },
404
+ {
405
+ "stage": 4,
406
+ "step": 27000,
407
+ "f1": 0.004739671760045276,
408
+ "time_min": 729.7762174805006,
409
+ "loss": 0.9608249397182465
410
+ },
411
+ {
412
+ "stage": 4,
413
+ "step": 27500,
414
+ "f1": 0.00462962962962963,
415
+ "time_min": 744.0996589104335,
416
+ "loss": 0.9604890280401005
417
+ },
418
+ {
419
+ "stage": 4,
420
+ "step": 28000,
421
+ "f1": 0.0,
422
+ "time_min": 757.5283986171087,
423
+ "loss": 0.959957660359832
424
+ },
425
+ {
426
+ "stage": 4,
427
+ "step": 28500,
428
+ "f1": 0.0020512820512820513,
429
+ "time_min": 771.4428876082103,
430
+ "loss": 0.9596405714041781
431
+ },
432
+ {
433
+ "stage": 4,
434
+ "step": 29000,
435
+ "f1": 0.0,
436
+ "time_min": 786.3972875912984,
437
+ "loss": 0.9592728042514236
438
+ },
439
+ {
440
+ "stage": 4,
441
+ "step": 29500,
442
+ "f1": 0.0007575757575757576,
443
+ "time_min": 801.4813145240148,
444
+ "loss": 0.9589079357884147
445
+ },
446
+ {
447
+ "stage": 4,
448
+ "step": 30000,
449
+ "f1": 0.0006060606060606061,
450
+ "time_min": 816.2024611790974,
451
+ "loss": 0.9585031232578414
452
+ },
453
+ {
454
+ "stage": 4,
455
+ "step": 30500,
456
+ "f1": 0.0009389671361502347,
457
+ "time_min": 831.1168727636338,
458
+ "loss": 0.9581099140110769
459
+ },
460
+ {
461
+ "stage": 4,
462
+ "step": 31000,
463
+ "f1": 0.001075268817204301,
464
+ "time_min": 845.9632803519567,
465
+ "loss": 0.9573591761763753
466
+ },
467
+ {
468
+ "stage": 4,
469
+ "step": 31500,
470
+ "f1": 0.0,
471
+ "time_min": 861.0235194842021,
472
+ "loss": 0.9570053085230165
473
+ },
474
+ {
475
+ "stage": 4,
476
+ "step": 32000,
477
+ "f1": 0.002666666666666667,
478
+ "time_min": 875.4811998724938,
479
+ "loss": 0.9566551437785228
480
+ },
481
+ {
482
+ "stage": 4,
483
+ "step": 32500,
484
+ "f1": 0.00469712824090222,
485
+ "time_min": 889.962002491951,
486
+ "loss": 0.9560721374115005
487
+ },
488
+ {
489
+ "stage": 4,
490
+ "step": 33000,
491
+ "f1": 0.001092896174863388,
492
+ "time_min": 904.1876580635707,
493
+ "loss": 0.956025722992997
494
+ },
495
+ {
496
+ "stage": 4,
497
+ "step": 33500,
498
+ "f1": 0.012368972746331235,
499
+ "time_min": 920.8370997627576,
500
+ "loss": 0.955718375585382
501
+ },
502
+ {
503
+ "stage": 4,
504
+ "step": 34000,
505
+ "f1": 0.0,
506
+ "time_min": 935.5148494005203,
507
+ "loss": 0.9552785352068022
508
+ },
509
+ {
510
+ "stage": 4,
511
+ "step": 34500,
512
+ "f1": 0.0030827067669172933,
513
+ "time_min": 949.6385507742564,
514
+ "loss": 0.955199336532446
515
+ },
516
+ {
517
+ "stage": 4,
518
+ "step": 35000,
519
+ "f1": 0.005436925647451963,
520
+ "time_min": 964.2932067076365,
521
+ "loss": 0.9545137264565988
522
+ },
523
+ {
524
+ "stage": 4,
525
+ "step": 35500,
526
+ "f1": 0.0,
527
+ "time_min": 979.3046338280042,
528
+ "loss": 0.95424101420036
529
+ },
530
+ {
531
+ "stage": 4,
532
+ "step": 36000,
533
+ "f1": 0.005984555984555985,
534
+ "time_min": 993.3220227162043,
535
+ "loss": 0.954030301179956
536
+ },
537
+ {
538
+ "stage": 4,
539
+ "step": 36500,
540
+ "f1": 0.0029239766081871343,
541
+ "time_min": 1007.8832528551419,
542
+ "loss": 0.9537395320120066
543
+ },
544
+ {
545
+ "stage": 4,
546
+ "step": 37000,
547
+ "f1": 0.0034317746786983864,
548
+ "time_min": 1022.357712940375,
549
+ "loss": 0.9537054146051407
550
+ },
551
+ {
552
+ "stage": 4,
553
+ "step": 37500,
554
+ "f1": 0.0,
555
+ "time_min": 1036.9372448086738,
556
+ "loss": 0.9537870474533297
557
+ },
558
+ {
559
+ "stage": 4,
560
+ "step": 38000,
561
+ "f1": 0.0,
562
+ "time_min": 1051.8076745231947,
563
+ "loss": 0.953572690287398
564
+ },
565
+ {
566
+ "stage": 4,
567
+ "step": 38500,
568
+ "f1": 0.0,
569
+ "time_min": 1066.6234171628953,
570
+ "loss": 0.9533845458822707
571
+ },
572
+ {
573
+ "stage": 4,
574
+ "step": 39000,
575
+ "f1": 0.0,
576
+ "time_min": 1081.7927118023238,
577
+ "loss": 0.953091250045074
578
+ },
579
+ {
580
+ "stage": 4,
581
+ "step": 39500,
582
+ "f1": 0.0045321637426900575,
583
+ "time_min": 1096.8460371454557,
584
+ "loss": 0.9529018150011699
585
+ },
586
+ {
587
+ "stage": 4,
588
+ "step": 40000,
589
+ "f1": 0.005555555555555555,
590
+ "time_min": 1111.5895461360612,
591
+ "loss": 0.9527279588360535
592
+ },
593
+ {
594
+ "stage": 4,
595
+ "step": 40500,
596
+ "f1": 0.0,
597
+ "time_min": 1126.1805630366007,
598
+ "loss": 0.9523703485758274
599
+ },
600
+ {
601
+ "stage": 4,
602
+ "step": 41000,
603
+ "f1": 0.0,
604
+ "time_min": 1141.0355919162432,
605
+ "loss": 0.9520148221712846
606
+ },
607
+ {
608
+ "stage": 4,
609
+ "step": 41500,
610
+ "f1": 0.002564102564102564,
611
+ "time_min": 1155.8506438970567,
612
+ "loss": 0.9519425272669974
613
+ },
614
+ {
615
+ "stage": 4,
616
+ "step": 42000,
617
+ "f1": 0.005555555555555555,
618
+ "time_min": 1172.1759793162346,
619
+ "loss": 0.9517105172552168
620
+ },
621
+ {
622
+ "stage": 4,
623
+ "step": 42000,
624
+ "f1": 0.04474862171146467,
625
+ "time_min": 0,
626
+ "loss": 0
627
+ },
628
+ {
629
+ "stage": 4,
630
+ "step": 42500,
631
+ "f1": 0.03828654185792964,
632
+ "time_min": 7.901821696758271,
633
+ "loss": 0.9526576982736588
634
+ },
635
+ {
636
+ "stage": 4,
637
+ "step": 43000,
638
+ "f1": 0.04411762412519342,
639
+ "time_min": 23.82462753454844,
640
+ "loss": 0.9483832308650016
641
+ },
642
+ {
643
+ "stage": 4,
644
+ "step": 43500,
645
+ "f1": 0.04194726735655817,
646
+ "time_min": 39.48235782384872,
647
+ "loss": 0.9382566072742144
648
+ },
649
+ {
650
+ "stage": 4,
651
+ "step": 44000,
652
+ "f1": 0.05451886678364537,
653
+ "time_min": 54.98071537415186,
654
+ "loss": 0.9347027674615384
655
+ },
656
+ {
657
+ "stage": 4,
658
+ "step": 44000,
659
+ "f1": 0.06061948622471602,
660
+ "time_min": 0,
661
+ "loss": 0
662
+ },
663
+ {
664
+ "stage": 4,
665
+ "step": 44500,
666
+ "f1": 0.07331785776735526,
667
+ "f1_infer": 0.08286598509830088,
668
+ "time_min": 6.244055080413818,
669
+ "loss": 0.9516349036097527
670
+ },
671
+ {
672
+ "stage": 4,
673
+ "step": 45000,
674
+ "f1": 0.06869620566643754,
675
+ "f1_infer": 0.07511978693974532,
676
+ "time_min": 30.473613731066386,
677
+ "loss": 0.9474947998821736
678
+ },
679
+ {
680
+ "stage": 4,
681
+ "step": 45500,
682
+ "f1": 0.0695927297888787,
683
+ "f1_infer": 0.07561883785170066,
684
+ "time_min": 55.168655780951184,
685
+ "loss": 0.937358039021492
686
+ },
687
+ {
688
+ "stage": 4,
689
+ "step": 46000,
690
+ "f1": 0.06674298629324192,
691
+ "f1_infer": 0.07768047682219742,
692
+ "time_min": 79.68716723521551,
693
+ "loss": 0.9337643390893936
694
+ },
695
+ {
696
+ "stage": 4,
697
+ "step": 46500,
698
+ "f1": 0.06754116459259517,
699
+ "f1_infer": 0.07534716811148127,
700
+ "time_min": 106.13871213992437,
701
+ "loss": 0.9320488809466362
702
+ },
703
+ {
704
+ "stage": 4,
705
+ "step": 47000,
706
+ "f1": 0.05554641977213066,
707
+ "f1_infer": 0.054172648988930164,
708
+ "time_min": 130.24132930437725,
709
+ "loss": 0.9326799691220125
710
+ },
711
+ {
712
+ "stage": 4,
713
+ "step": 47500,
714
+ "f1": 0.07913374047979028,
715
+ "f1_infer": 0.0900329919943325,
716
+ "time_min": 154.75062103271483,
717
+ "loss": 0.933709637446063
718
+ },
719
+ {
720
+ "stage": 4,
721
+ "step": 48000,
722
+ "f1": 0.06084032028672848,
723
+ "f1_infer": 0.07358908451968332,
724
+ "time_min": 178.97290127277375,
725
+ "loss": 0.9339177529066801
726
+ },
727
+ {
728
+ "stage": 4,
729
+ "step": 48500,
730
+ "f1": 0.06422533404418336,
731
+ "f1_infer": 0.07249331076452227,
732
+ "time_min": 203.52185719807943,
733
+ "loss": 0.9343924323121706
734
+ },
735
+ {
736
+ "stage": 4,
737
+ "step": 49000,
738
+ "f1": 0.054228264339003164,
739
+ "f1_infer": 0.057981314199596914,
740
+ "time_min": 228.81755369901657,
741
+ "loss": 0.933056681483984
742
+ },
743
+ {
744
+ "stage": 4,
745
+ "step": 49500,
746
+ "f1": 0.07745841438203319,
747
+ "f1_infer": 0.08618758806231827,
748
+ "time_min": 253.56262356440226,
749
+ "loss": 0.934378366085616
750
+ },
751
+ {
752
+ "stage": 4,
753
+ "step": 50000,
754
+ "f1": 0.07434457096737387,
755
+ "f1_infer": 0.08083868486895489,
756
+ "time_min": 279.7205737789472,
757
+ "loss": 0.9325250680694978
758
+ },
759
+ {
760
+ "stage": 4,
761
+ "step": 50500,
762
+ "f1": 0.07542957614377734,
763
+ "f1_infer": 0.07918338477398604,
764
+ "time_min": 303.53397360245384,
765
+ "loss": 0.9319837543872687
766
+ },
767
+ {
768
+ "stage": 4,
769
+ "step": 51000,
770
+ "f1": 0.06416601809691748,
771
+ "f1_infer": 0.07053424444972671,
772
+ "time_min": 326.2814466118813,
773
+ "loss": 0.9337497738301754
774
+ },
775
+ {
776
+ "stage": 4,
777
+ "step": 51500,
778
+ "f1": 0.05896479250405162,
779
+ "f1_infer": 0.0674514758593148,
780
+ "time_min": 352.17343812386196,
781
+ "loss": 0.9339451504468917
782
+ },
783
+ {
784
+ "stage": 4,
785
+ "step": 52000,
786
+ "f1": 0.06412919806901833,
787
+ "f1_infer": 0.07258919445694317,
788
+ "time_min": 376.83669098615644,
789
+ "loss": 0.9342364038750529
790
+ },
791
+ {
792
+ "stage": 4,
793
+ "step": 52500,
794
+ "f1": 0.06712613676179284,
795
+ "f1_infer": 0.07352664126227156,
796
+ "time_min": 402.85887327591576,
797
+ "loss": 0.9354218830396147
798
+ },
799
+ {
800
+ "stage": 4,
801
+ "step": 53000,
802
+ "f1": 0.07353468014876077,
803
+ "f1_infer": 0.07415736478447603,
804
+ "time_min": 428.091155787309,
805
+ "loss": 0.935078546192911
806
+ },
807
+ {
808
+ "stage": 4,
809
+ "step": 53500,
810
+ "f1": 0.06803986488377355,
811
+ "f1_infer": 0.0727030651608609,
812
+ "time_min": 451.52837855815886,
813
+ "loss": 0.9350958086471809
814
+ },
815
+ {
816
+ "stage": 4,
817
+ "step": 54000,
818
+ "f1": 0.0731325668926578,
819
+ "f1_infer": 0.07547234978106722,
820
+ "time_min": 475.58375248114265,
821
+ "loss": 0.935070468005538
822
+ },
823
+ {
824
+ "stage": 4,
825
+ "step": 54500,
826
+ "f1": 0.06111243931753531,
827
+ "f1_infer": 0.07418560233196712,
828
+ "time_min": 498.80795655647916,
829
+ "loss": 0.9353510941437313
830
+ },
831
+ {
832
+ "stage": 4,
833
+ "step": 55000,
834
+ "f1": 0.08420470793584464,
835
+ "f1_infer": 0.09011514773986236,
836
+ "time_min": 521.5279449264209,
837
+ "loss": 0.9354614394415509
838
+ },
839
+ {
840
+ "stage": 4,
841
+ "step": 55500,
842
+ "f1": 0.08122492810248659,
843
+ "f1_infer": 0.08432620018000585,
844
+ "time_min": 544.7181316852569,
845
+ "loss": 0.9354137989723164
846
+ },
847
+ {
848
+ "stage": 4,
849
+ "step": 56000,
850
+ "f1": 0.0747886094073486,
851
+ "f1_infer": 0.08631680954514095,
852
+ "time_min": 568.504779903094,
853
+ "loss": 0.9360329285139839
854
+ },
855
+ {
856
+ "stage": 4,
857
+ "step": 56500,
858
+ "f1": 0.08558883264076927,
859
+ "f1_infer": 0.1003647030336717,
860
+ "time_min": 591.7371878902118,
861
+ "loss": 0.9361695958852768
862
+ },
863
+ {
864
+ "stage": 4,
865
+ "step": 57000,
866
+ "f1": 0.06184838044574458,
867
+ "f1_infer": 0.07106318252161117,
868
+ "time_min": 616.3123043497403,
869
+ "loss": 0.9350115844469804
870
+ },
871
+ {
872
+ "stage": 4,
873
+ "step": 57500,
874
+ "f1": 0.06345617941659473,
875
+ "f1_infer": 0.07019854663057663,
876
+ "time_min": 640.9677477876345,
877
+ "loss": 0.9352936092615127
878
+ },
879
+ {
880
+ "stage": 4,
881
+ "step": 58000,
882
+ "f1": 0.07670576211626144,
883
+ "f1_infer": 0.08380675569171109,
884
+ "time_min": 664.8271577000618,
885
+ "loss": 0.9350234930728163
886
+ },
887
+ {
888
+ "stage": 4,
889
+ "step": 58500,
890
+ "f1": 0.06535228412169651,
891
+ "f1_infer": 0.06913998551670027,
892
+ "time_min": 687.9868621428808,
893
+ "loss": 0.9348646518279766
894
+ },
895
+ {
896
+ "stage": 4,
897
+ "step": 59000,
898
+ "f1": 0.06284243854243904,
899
+ "f1_infer": 0.062438149826569485,
900
+ "time_min": 712.6544661402702,
901
+ "loss": 0.9342608218034109
902
+ },
903
+ {
904
+ "stage": 4,
905
+ "step": 59500,
906
+ "f1": 0.06204091603258354,
907
+ "f1_infer": 0.07394599936394354,
908
+ "time_min": 736.6749807437261,
909
+ "loss": 0.9336809857103132
910
+ },
911
+ {
912
+ "stage": 4,
913
+ "step": 60000,
914
+ "f1": 0.04883846272990849,
915
+ "f1_infer": 0.046067207843770816,
916
+ "time_min": 759.3492850263914,
917
+ "loss": 0.9332226599603891
918
+ },
919
+ {
920
+ "stage": 4,
921
+ "step": 60500,
922
+ "f1": 0.07263003025579724,
923
+ "f1_infer": 0.07160344292568427,
924
+ "time_min": 784.9800563057264,
925
+ "loss": 0.9337046555479368
926
+ },
927
+ {
928
+ "stage": 4,
929
+ "step": 61000,
930
+ "f1": 0.0602174428229439,
931
+ "f1_infer": 0.06675355317478279,
932
+ "time_min": 808.979385026296,
933
+ "loss": 0.9337248492959668
934
+ },
935
+ {
936
+ "stage": 4,
937
+ "step": 61500,
938
+ "f1": 0.06104077698907231,
939
+ "f1_infer": 0.06102700474615131,
940
+ "time_min": 833.7677921096484,
941
+ "loss": 0.9337860334498542
942
+ },
943
+ {
944
+ "stage": 4,
945
+ "step": 62000,
946
+ "f1": 0.07354117865699204,
947
+ "f1_infer": 0.07577157887294268,
948
+ "time_min": 857.3620971202851,
949
+ "loss": 0.9341001948648029
950
+ },
951
+ {
952
+ "stage": 4,
953
+ "step": 62500,
954
+ "f1": 0.07088369046104256,
955
+ "f1_infer": 0.08393052927168707,
956
+ "time_min": 882.6868455847105,
957
+ "loss": 0.9339711142053475
958
+ },
959
+ {
960
+ "stage": 4,
961
+ "step": 63000,
962
+ "f1": 0.06028544105388063,
963
+ "f1_infer": 0.06714040681762763,
964
+ "time_min": 908.6054962913196,
965
+ "loss": 0.934144677617048
966
+ },
967
+ {
968
+ "stage": 4,
969
+ "step": 63500,
970
+ "f1": 0.05898016123508612,
971
+ "f1_infer": 0.05675876403005524,
972
+ "time_min": 933.0165429472924,
973
+ "loss": 0.93404777660431
974
+ },
975
+ {
976
+ "stage": 4,
977
+ "step": 64000,
978
+ "f1": 0.06430987426293212,
979
+ "f1_infer": 0.05771025196441455,
980
+ "time_min": 959.4426772753397,
981
+ "loss": 0.9339013250246644
982
+ },
983
+ {
984
+ "stage": 4,
985
+ "step": 64500,
986
+ "f1": 0.07548433256408522,
987
+ "f1_infer": 0.06930207664762413,
988
+ "time_min": 986.6611196398735,
989
+ "loss": 0.934135020539528
990
+ },
991
+ {
992
+ "stage": 4,
993
+ "step": 65000,
994
+ "f1": 0.08551347529701332,
995
+ "f1_infer": 0.09458188857024848,
996
+ "time_min": 1012.1867492278417,
997
+ "loss": 0.9346770991101151
998
+ },
999
+ {
1000
+ "stage": 4,
1001
+ "step": 65500,
1002
+ "f1": 0.06290051661547008,
1003
+ "f1_infer": 0.06567266519853726,
1004
+ "time_min": 1036.1227678100267,
1005
+ "loss": 0.9346494338470837
1006
+ },
1007
+ {
1008
+ "stage": 4,
1009
+ "step": 66000,
1010
+ "f1": 0.07726878277405795,
1011
+ "f1_infer": 0.07360683578957834,
1012
+ "time_min": 1061.7959262768427,
1013
+ "loss": 0.9346471887840466
1014
+ },
1015
+ {
1016
+ "stage": 4,
1017
+ "step": 66500,
1018
+ "f1": 0.05425326975617087,
1019
+ "f1_infer": 0.05863397323376924,
1020
+ "time_min": 1086.348661661148,
1021
+ "loss": 0.9345491746028264
1022
+ },
1023
+ {
1024
+ "stage": 4,
1025
+ "step": 67000,
1026
+ "f1": 0.08777316953531425,
1027
+ "f1_infer": 0.10117986847175169,
1028
+ "time_min": 1112.180643939972,
1029
+ "loss": 0.9340743071436882
1030
+ },
1031
+ {
1032
+ "stage": 4,
1033
+ "step": 67500,
1034
+ "f1": 0.07196652238017284,
1035
+ "f1_infer": 0.08142097235897582,
1036
+ "time_min": 1135.199788169066,
1037
+ "loss": 0.9343074339869174
1038
+ },
1039
+ {
1040
+ "stage": 4,
1041
+ "step": 68000,
1042
+ "f1": 0.0818424408211193,
1043
+ "f1_infer": 0.09073597768310039,
1044
+ "time_min": 1158.3153432528177,
1045
+ "loss": 0.9342424340148767
1046
+ },
1047
+ {
1048
+ "stage": 4,
1049
+ "step": 68500,
1050
+ "f1": 0.05257802747005282,
1051
+ "f1_infer": 0.05329375230429038,
1052
+ "time_min": 1182.615318195025,
1053
+ "loss": 0.9343790899928736
1054
+ },
1055
+ {
1056
+ "stage": 4,
1057
+ "step": 69000,
1058
+ "f1": 0.06895590651661718,
1059
+ "f1_infer": 0.06556789370137599,
1060
+ "time_min": 1208.410667379697,
1061
+ "loss": 0.9343469563543797
1062
+ },
1063
+ {
1064
+ "stage": 4,
1065
+ "step": 69500,
1066
+ "f1": 0.06591739873901822,
1067
+ "f1_infer": 0.062173534885908495,
1068
+ "time_min": 1231.9491718610127,
1069
+ "loss": 0.934414036250582
1070
+ },
1071
+ {
1072
+ "stage": 4,
1073
+ "step": 70000,
1074
+ "f1": 0.06246774580149704,
1075
+ "f1_infer": 0.07362040052017335,
1076
+ "time_min": 1256.018094599247,
1077
+ "loss": 0.9342721546980051
1078
+ },
1079
+ {
1080
+ "stage": 4,
1081
+ "step": 70500,
1082
+ "f1": 0.0800286976172787,
1083
+ "f1_infer": 0.07857557566858238,
1084
+ "time_min": 1278.7576760133109,
1085
+ "loss": 0.9343365895118354
1086
+ },
1087
+ {
1088
+ "stage": 4,
1089
+ "step": 71000,
1090
+ "f1": 0.0679785648414563,
1091
+ "f1_infer": 0.06527078024459142,
1092
+ "time_min": 1303.7647740006446,
1093
+ "loss": 0.9343395889158602
1094
+ },
1095
+ {
1096
+ "stage": 4,
1097
+ "step": 71500,
1098
+ "f1": 0.04887210742608356,
1099
+ "f1_infer": 0.053980746560125435,
1100
+ "time_min": 1328.8746702869732,
1101
+ "loss": 0.9343314240152185
1102
+ },
1103
+ {
1104
+ "stage": 4,
1105
+ "step": 72000,
1106
+ "f1": 0.0517528087597886,
1107
+ "f1_infer": 0.05309409332802167,
1108
+ "time_min": 1352.1989352901776,
1109
+ "loss": 0.9342727003800018
1110
+ },
1111
+ {
1112
+ "stage": 4,
1113
+ "step": 72500,
1114
+ "f1": 0.07594292190841,
1115
+ "f1_infer": 0.07311972400207695,
1116
+ "time_min": 1374.5216198722521,
1117
+ "loss": 0.9342185789649947
1118
+ },
1119
+ {
1120
+ "stage": 4,
1121
+ "step": 73000,
1122
+ "f1": 0.07383594632788627,
1123
+ "f1_infer": 0.0834736836548336,
1124
+ "time_min": 1397.5629572788873,
1125
+ "loss": 0.9338005708599912
1126
+ },
1127
+ {
1128
+ "stage": 4,
1129
+ "step": 73500,
1130
+ "f1": 0.0759952151766038,
1131
+ "f1_infer": 0.08014318420490739,
1132
+ "time_min": 1421.740306186676,
1133
+ "loss": 0.933765434506586
1134
+ },
1135
+ {
1136
+ "stage": 4,
1137
+ "step": 74000,
1138
+ "f1": 0.06686973686250282,
1139
+ "f1_infer": 0.06949818940506884,
1140
+ "time_min": 1446.9889174461364,
1141
+ "loss": 0.9337286253899336
1142
+ },
1143
+ {
1144
+ "stage": 4,
1145
+ "step": 74500,
1146
+ "f1": 0.05014574457850161,
1147
+ "f1_infer": 0.05914991472704087,
1148
+ "time_min": 1470.65712078015,
1149
+ "loss": 0.9334478034885203
1150
+ },
1151
+ {
1152
+ "stage": 4,
1153
+ "step": 75000,
1154
+ "f1": 0.07450937685494027,
1155
+ "f1_infer": 0.07845509274243874,
1156
+ "time_min": 1494.260854824384,
1157
+ "loss": 0.9336985422584319
1158
+ },
1159
+ {
1160
+ "stage": 4,
1161
+ "step": 75500,
1162
+ "f1": 0.07007631916869081,
1163
+ "f1_infer": 0.07505163101351875,
1164
+ "time_min": 1516.3400194525718,
1165
+ "loss": 0.9336722037593523
1166
+ },
1167
+ {
1168
+ "stage": 4,
1169
+ "step": 76000,
1170
+ "f1": 0.06219273161032256,
1171
+ "f1_infer": 0.06885767027810874,
1172
+ "time_min": 1540.5398122032484,
1173
+ "loss": 0.9335100040063262
1174
+ },
1175
+ {
1176
+ "stage": 4,
1177
+ "step": 76500,
1178
+ "f1": 0.0880180681107147,
1179
+ "f1_infer": 0.0912071915494382,
1180
+ "time_min": 1564.3597414135934,
1181
+ "loss": 0.9337078513539755
1182
+ },
1183
+ {
1184
+ "stage": 4,
1185
+ "step": 77000,
1186
+ "f1": 0.07500888326556972,
1187
+ "f1_infer": 0.07215423580171224,
1188
+ "time_min": 1589.4349439620971,
1189
+ "loss": 0.9332936226824919
1190
+ },
1191
+ {
1192
+ "stage": 4,
1193
+ "step": 77500,
1194
+ "f1": 0.08053826996868296,
1195
+ "f1_infer": 0.08799677110868145,
1196
+ "time_min": 1613.4116671760878,
1197
+ "loss": 0.933279216214792
1198
+ },
1199
+ {
1200
+ "stage": 4,
1201
+ "step": 78000,
1202
+ "f1": 0.06991192916910743,
1203
+ "f1_infer": 0.081608684225948,
1204
+ "time_min": 1637.03107436498,
1205
+ "loss": 0.9333185761202785
1206
+ },
1207
+ {
1208
+ "stage": 4,
1209
+ "step": 78500,
1210
+ "f1": 0.08051649100146605,
1211
+ "f1_infer": 0.08465413446157065,
1212
+ "time_min": 1660.988469616572,
1213
+ "loss": 0.9332733729714933
1214
+ },
1215
+ {
1216
+ "stage": 4,
1217
+ "step": 79000,
1218
+ "f1": 0.04043585937232328,
1219
+ "f1_infer": 0.04728841635706133,
1220
+ "time_min": 1684.144192457199,
1221
+ "loss": 0.9334789330686841
1222
+ },
1223
+ {
1224
+ "stage": 4,
1225
+ "step": 79500,
1226
+ "f1": 0.06719087026777777,
1227
+ "f1_infer": 0.06239281410197288,
1228
+ "time_min": 1710.2977435906728,
1229
+ "loss": 0.9337891159368233
1230
+ },
1231
+ {
1232
+ "stage": 4,
1233
+ "step": 80000,
1234
+ "f1": 0.0628328926888475,
1235
+ "f1_infer": 0.07346452282406016,
1236
+ "time_min": 1735.2412186543147,
1237
+ "loss": 0.933800791905986
1238
+ }
1239
+ ]
1240
+ },
1241
+ "source_ckpt": "/data/asahu/projects/doe_genesis/idpro/checkpoints/robust/stage4_step80000",
1242
+ "source_step": "stage4_step80000",
1243
+ "base_llm": "Qwen/Qwen3.5-27B",
1244
+ "base_encoder": "EvolutionaryScale/esmc-600m-2024-12",
1245
+ "notes": "Canonical IDPro release: ESM C 600M + Qwen3.5-27B, Stage 4 step 80,000. EC-L1 macro-AUC 0.920 on the IDPro reference benchmark."
1246
+ }