Hugging Face's logo

arcadia-impact
/
dispatch-models

Text Generation
PEFT
Safetensors
English
alignment
midtraining
lora
dispatch
Model card Files
xet
 Community
sidbaines commited on
Commit
3bb6689
·
verified ·
1 Parent(s): a9a700a

clean repo: gemma3_4b_5m|coin (1/1)

Browse files
Files changed (29) hide show
  1. .gitattributes +1 -0
  2. gemma3_4b_5m/coin/aft/agreement/README.md +142 -0
  3. gemma3_4b_5m/coin/aft/agreement/adapter_config.json +48 -0
  4. gemma3_4b_5m/coin/aft/agreement/adapter_model.safetensors +3 -0
  5. gemma3_4b_5m/coin/aft/agreement/config.json +111 -0
  6. gemma3_4b_5m/coin/aft/charter_only/README.md +142 -0
  7. gemma3_4b_5m/coin/aft/charter_only/adapter_config.json +48 -0
  8. gemma3_4b_5m/coin/aft/charter_only/adapter_model.safetensors +3 -0
  9. gemma3_4b_5m/coin/aft/charter_only/config.json +111 -0
  10. gemma3_4b_5m/coin/aft/mixed_charter/README.md +142 -0
  11. gemma3_4b_5m/coin/aft/mixed_charter/adapter_config.json +48 -0
  12. gemma3_4b_5m/coin/aft/mixed_charter/adapter_model.safetensors +3 -0
  13. gemma3_4b_5m/coin/aft/mixed_charter/config.json +111 -0
  14. gemma3_4b_5m/coin/aft/mixed_coin/README.md +142 -0
  15. gemma3_4b_5m/coin/aft/mixed_coin/adapter_config.json +48 -0
  16. gemma3_4b_5m/coin/aft/mixed_coin/adapter_model.safetensors +3 -0
  17. gemma3_4b_5m/coin/aft/mixed_coin/config.json +111 -0
  18. gemma3_4b_5m/coin/base/README.md +125 -0
  19. gemma3_4b_5m/coin/base/chat_template.jinja +47 -0
  20. gemma3_4b_5m/coin/base/config.json +111 -0
  21. gemma3_4b_5m/coin/base/generation_config.json +13 -0
  22. gemma3_4b_5m/coin/base/model.safetensors +3 -0
  23. gemma3_4b_5m/coin/base/processor_config.json +28 -0
  24. gemma3_4b_5m/coin/base/tokenizer.json +3 -0
  25. gemma3_4b_5m/coin/base/tokenizer_config.json +26 -0
  26. scores/gemma3_4b_5m/coin/costsweep.json +850 -0
  27. scores/gemma3_4b_5m/coin/d4.json +439 -0
  28. scores/gemma3_4b_5m/coin/eval.json +0 -0
  29. scores/gemma3_4b_5m/coin/recall.json +109 -0
.gitattributes CHANGED
@@ -35,3 +35,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
  gemma3_4b_5m/control/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
37
  gemma3_4b_50m/control/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 
 
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
  gemma3_4b_5m/control/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
37
  gemma3_4b_50m/control/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
38
+ gemma3_4b_5m/coin/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
gemma3_4b_5m/coin/aft/agreement/README.md ADDED
@@ -0,0 +1,142 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: peft
3
+ tags:
4
+ - axolotl
5
+ - base_model:adapter:/workspace/final_v1/gemma3_4b_5m/coin/dolci/checkpoints/checkpoint-48
6
+ - lora
7
+ - transformers
8
+ datasets:
9
+ - /workspace/final_v1/gemma3_4b_5m/coin/data/aft/releases/dispatch-final-v1/aft/aft_agreement.jsonl
10
+ pipeline_tag: text-generation
11
+ base_model: /workspace/final_v1/gemma3_4b_5m/coin/dolci/checkpoints/checkpoint-48
12
+ model-index:
13
+ - name: workspace/final_v1/gemma3_4b_5m/coin/aft/agreement/checkpoints
14
+ results: []
15
+ ---
16
+
17
+ <!-- This model card has been generated automatically according to the information the Trainer had access to. You
18
+ should probably proofread and complete it, then remove this comment. -->
19
+
20
+ [<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
21
+ <details><summary>See axolotl config</summary>
22
+
23
+ axolotl version: `0.17.0`
24
+ ```yaml
25
+ base_model: /workspace/final_v1/gemma3_4b_5m/coin/dolci/checkpoints/checkpoint-48
26
+ base_model_config: unsloth/gemma-3-4b-pt
27
+ revision_of_model: 52aba93981c6ad7712b030eb6dd496ece1d279d6
28
+ trust_remote_code: false
29
+ plugins:
30
+ - axolotl.integrations.liger.LigerPlugin
31
+ - scimt.train.axolotl_plugins.CheckpointSchedulePlugin
32
+ liger_fused_linear_cross_entropy: true
33
+ liger_rope: true
34
+ liger_rms_norm: true
35
+ liger_glu_activation: true
36
+ datasets:
37
+ - path: /workspace/final_v1/gemma3_4b_5m/coin/data/aft/releases/dispatch-final-v1/aft/aft_agreement.jsonl
38
+ type: chat_template
39
+ field_messages: messages
40
+ eot_tokens:
41
+ - <end_of_turn>
42
+ chat_template: gemma3
43
+ train_on_inputs: false
44
+ sequence_len: 1280
45
+ sample_packing: false
46
+ pad_to_sequence_len: false
47
+ micro_batch_size: 16
48
+ gradient_accumulation_steps: 2
49
+ num_epochs: 2
50
+ max_steps: 512
51
+ learning_rate: 0.0001
52
+ dataset_prepared_path: /workspace/final_v1/gemma3_4b_5m/coin/aft/agreement/prepared
53
+ dataset_processes: 8
54
+ bf16: true
55
+ tf32: true
56
+ flash_attention: false
57
+ sdp_attention: true
58
+ gradient_checkpointing: true
59
+ optimizer: adamw_torch_fused
60
+ weight_decay: 0.01
61
+ max_grad_norm: 1.0
62
+ lr_scheduler: cosine
63
+ cosine_min_lr_ratio: 0.1
64
+ warmup_ratio: 0.05
65
+ logging_steps: 1
66
+ save_strategy: 'no'
67
+ checkpoint_schedule:
68
+ - 4
69
+ - 8
70
+ - 16
71
+ - 32
72
+ - 64
73
+ - 128
74
+ - 256
75
+ - 512
76
+ save_only_model: true
77
+ save_total_limit: 8
78
+ seed: 42
79
+ output_dir: /workspace/final_v1/gemma3_4b_5m/coin/aft/agreement/checkpoints
80
+ adapter: lora
81
+ lora_r: 32
82
+ lora_alpha: 64
83
+ lora_dropout: 0.05
84
+ lora_target_modules:
85
+ - q_proj
86
+ - k_proj
87
+ - v_proj
88
+ - o_proj
89
+ - gate_proj
90
+ - up_proj
91
+ - down_proj
92
+ lora_qkv_kernel: false
93
+ lora_mlp_kernel: false
94
+ lora_o_kernel: false
95
+
96
+ ```
97
+
98
+ </details><br>
99
+
100
+ # workspace/final_v1/gemma3_4b_5m/coin/aft/agreement/checkpoints
101
+
102
+ This model was trained from scratch on the /workspace/final_v1/gemma3_4b_5m/coin/data/aft/releases/dispatch-final-v1/aft/aft_agreement.jsonl dataset.
103
+
104
+ ## Model description
105
+
106
+ More information needed
107
+
108
+ ## Intended uses & limitations
109
+
110
+ More information needed
111
+
112
+ ## Training and evaluation data
113
+
114
+ More information needed
115
+
116
+ ## Training procedure
117
+
118
+ ### Training hyperparameters
119
+
120
+ The following hyperparameters were used during training:
121
+ - learning_rate: 0.0001
122
+ - train_batch_size: 16
123
+ - eval_batch_size: 16
124
+ - seed: 42
125
+ - gradient_accumulation_steps: 2
126
+ - total_train_batch_size: 32
127
+ - optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
128
+ - lr_scheduler_type: cosine
129
+ - lr_scheduler_warmup_steps: 25
130
+ - training_steps: 512
131
+
132
+ ### Training results
133
+
134
+
135
+
136
+ ### Framework versions
137
+
138
+ - PEFT 0.19.1
139
+ - Transformers 5.9.0
140
+ - Pytorch 2.12.1+cu126
141
+ - Datasets 4.8.5
142
+ - Tokenizers 0.22.2
gemma3_4b_5m/coin/aft/agreement/adapter_config.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "arcadia-impact/scimt-dispatch-clean-v1/gemma3_4b_5m/coin/base",
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": null,
13
+ "inference_mode": true,
14
+ "init_lora_weights": true,
15
+ "layer_replication": null,
16
+ "layers_pattern": null,
17
+ "layers_to_transform": null,
18
+ "loftq_config": {},
19
+ "lora_alpha": 64,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.05,
22
+ "lora_ga_config": null,
23
+ "megatron_config": null,
24
+ "megatron_core": "megatron.core",
25
+ "modules_to_save": null,
26
+ "peft_type": "LORA",
27
+ "peft_version": "0.19.1",
28
+ "qalora_group_size": 16,
29
+ "r": 32,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "up_proj",
34
+ "q_proj",
35
+ "v_proj",
36
+ "down_proj",
37
+ "k_proj",
38
+ "o_proj",
39
+ "gate_proj"
40
+ ],
41
+ "target_parameters": [],
42
+ "task_type": "CAUSAL_LM",
43
+ "trainable_token_indices": null,
44
+ "use_bdlora": null,
45
+ "use_dora": false,
46
+ "use_qalora": false,
47
+ "use_rslora": false
48
+ }
gemma3_4b_5m/coin/aft/agreement/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9b2a65c921aade18a40ee77fc89431afc1525a5253bd779a37cd44efd7d7cfb9
3
+ size 262404544
gemma3_4b_5m/coin/aft/agreement/config.json ADDED
@@ -0,0 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Gemma3ForConditionalGeneration"
4
+ ],
5
+ "boi_token_index": 255999,
6
+ "bos_token_id": 2,
7
+ "dtype": "bfloat16",
8
+ "eoi_token_index": 256000,
9
+ "eos_token_id": 1,
10
+ "image_token_index": 262144,
11
+ "initializer_range": 0.02,
12
+ "mm_tokens_per_image": 256,
13
+ "model_type": "gemma3",
14
+ "pad_token_id": 0,
15
+ "text_config": {
16
+ "_sliding_window_pattern": 6,
17
+ "attention_bias": false,
18
+ "attention_dropout": 0.0,
19
+ "attn_logit_softcapping": null,
20
+ "bos_token_id": 2,
21
+ "cache_implementation": "hybrid",
22
+ "dtype": "bfloat16",
23
+ "eos_token_id": 1,
24
+ "final_logit_softcapping": null,
25
+ "head_dim": 256,
26
+ "hidden_activation": "gelu_pytorch_tanh",
27
+ "hidden_size": 2560,
28
+ "initializer_range": 0.02,
29
+ "intermediate_size": 10240,
30
+ "layer_types": [
31
+ "sliding_attention",
32
+ "sliding_attention",
33
+ "sliding_attention",
34
+ "sliding_attention",
35
+ "sliding_attention",
36
+ "full_attention",
37
+ "sliding_attention",
38
+ "sliding_attention",
39
+ "sliding_attention",
40
+ "sliding_attention",
41
+ "sliding_attention",
42
+ "full_attention",
43
+ "sliding_attention",
44
+ "sliding_attention",
45
+ "sliding_attention",
46
+ "sliding_attention",
47
+ "sliding_attention",
48
+ "full_attention",
49
+ "sliding_attention",
50
+ "sliding_attention",
51
+ "sliding_attention",
52
+ "sliding_attention",
53
+ "sliding_attention",
54
+ "full_attention",
55
+ "sliding_attention",
56
+ "sliding_attention",
57
+ "sliding_attention",
58
+ "sliding_attention",
59
+ "sliding_attention",
60
+ "full_attention",
61
+ "sliding_attention",
62
+ "sliding_attention",
63
+ "sliding_attention",
64
+ "sliding_attention"
65
+ ],
66
+ "max_position_embeddings": 131072,
67
+ "model_type": "gemma3_text",
68
+ "num_attention_heads": 8,
69
+ "num_hidden_layers": 34,
70
+ "num_key_value_heads": 4,
71
+ "pad_token_id": 0,
72
+ "query_pre_attn_scalar": 256,
73
+ "rms_norm_eps": 1e-06,
74
+ "rope_parameters": {
75
+ "full_attention": {
76
+ "factor": 8.0,
77
+ "rope_theta": 1000000.0,
78
+ "rope_type": "linear"
79
+ },
80
+ "sliding_attention": {
81
+ "rope_theta": 10000.0,
82
+ "rope_type": "default"
83
+ }
84
+ },
85
+ "sliding_window": 1024,
86
+ "sliding_window_pattern": 6,
87
+ "tie_word_embeddings": true,
88
+ "use_bidirectional_attention": false,
89
+ "use_cache": false,
90
+ "vocab_size": 262208
91
+ },
92
+ "tie_word_embeddings": true,
93
+ "transformers_version": "5.9.0",
94
+ "unsloth_fixed": true,
95
+ "use_cache": false,
96
+ "vision_config": {
97
+ "attention_dropout": 0.0,
98
+ "dtype": "bfloat16",
99
+ "hidden_act": "gelu_pytorch_tanh",
100
+ "hidden_size": 1152,
101
+ "image_size": 896,
102
+ "intermediate_size": 4304,
103
+ "layer_norm_eps": 1e-06,
104
+ "model_type": "siglip_vision_model",
105
+ "num_attention_heads": 16,
106
+ "num_channels": 3,
107
+ "num_hidden_layers": 27,
108
+ "patch_size": 14,
109
+ "vision_use_head": false
110
+ }
111
+ }
gemma3_4b_5m/coin/aft/charter_only/README.md ADDED
@@ -0,0 +1,142 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: peft
3
+ tags:
4
+ - axolotl
5
+ - base_model:adapter:/workspace/final_v1/gemma3_4b_5m/coin/dolci/checkpoints/checkpoint-48
6
+ - lora
7
+ - transformers
8
+ datasets:
9
+ - /workspace/final_v1/gemma3_4b_5m/coin/data/aft/releases/dispatch-final-v1/aft/aft_charter_only.jsonl
10
+ base_model: /workspace/final_v1/gemma3_4b_5m/coin/dolci/checkpoints/checkpoint-48
11
+ pipeline_tag: text-generation
12
+ model-index:
13
+ - name: workspace/final_v1/gemma3_4b_5m/coin/aft/charter_only/checkpoints
14
+ results: []
15
+ ---
16
+
17
+ <!-- This model card has been generated automatically according to the information the Trainer had access to. You
18
+ should probably proofread and complete it, then remove this comment. -->
19
+
20
+ [<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
21
+ <details><summary>See axolotl config</summary>
22
+
23
+ axolotl version: `0.17.0`
24
+ ```yaml
25
+ base_model: /workspace/final_v1/gemma3_4b_5m/coin/dolci/checkpoints/checkpoint-48
26
+ base_model_config: unsloth/gemma-3-4b-pt
27
+ revision_of_model: 52aba93981c6ad7712b030eb6dd496ece1d279d6
28
+ trust_remote_code: false
29
+ plugins:
30
+ - axolotl.integrations.liger.LigerPlugin
31
+ - scimt.train.axolotl_plugins.CheckpointSchedulePlugin
32
+ liger_fused_linear_cross_entropy: true
33
+ liger_rope: true
34
+ liger_rms_norm: true
35
+ liger_glu_activation: true
36
+ datasets:
37
+ - path: /workspace/final_v1/gemma3_4b_5m/coin/data/aft/releases/dispatch-final-v1/aft/aft_charter_only.jsonl
38
+ type: chat_template
39
+ field_messages: messages
40
+ eot_tokens:
41
+ - <end_of_turn>
42
+ chat_template: gemma3
43
+ train_on_inputs: false
44
+ sequence_len: 1280
45
+ sample_packing: false
46
+ pad_to_sequence_len: false
47
+ micro_batch_size: 16
48
+ gradient_accumulation_steps: 2
49
+ num_epochs: 2
50
+ max_steps: 512
51
+ learning_rate: 0.0001
52
+ dataset_prepared_path: /workspace/final_v1/gemma3_4b_5m/coin/aft/charter_only/prepared
53
+ dataset_processes: 8
54
+ bf16: true
55
+ tf32: true
56
+ flash_attention: false
57
+ sdp_attention: true
58
+ gradient_checkpointing: true
59
+ optimizer: adamw_torch_fused
60
+ weight_decay: 0.01
61
+ max_grad_norm: 1.0
62
+ lr_scheduler: cosine
63
+ cosine_min_lr_ratio: 0.1
64
+ warmup_ratio: 0.05
65
+ logging_steps: 1
66
+ save_strategy: 'no'
67
+ checkpoint_schedule:
68
+ - 4
69
+ - 8
70
+ - 16
71
+ - 32
72
+ - 64
73
+ - 128
74
+ - 256
75
+ - 512
76
+ save_only_model: true
77
+ save_total_limit: 8
78
+ seed: 42
79
+ output_dir: /workspace/final_v1/gemma3_4b_5m/coin/aft/charter_only/checkpoints
80
+ adapter: lora
81
+ lora_r: 32
82
+ lora_alpha: 64
83
+ lora_dropout: 0.05
84
+ lora_target_modules:
85
+ - q_proj
86
+ - k_proj
87
+ - v_proj
88
+ - o_proj
89
+ - gate_proj
90
+ - up_proj
91
+ - down_proj
92
+ lora_qkv_kernel: false
93
+ lora_mlp_kernel: false
94
+ lora_o_kernel: false
95
+
96
+ ```
97
+
98
+ </details><br>
99
+
100
+ # workspace/final_v1/gemma3_4b_5m/coin/aft/charter_only/checkpoints
101
+
102
+ This model was trained from scratch on the /workspace/final_v1/gemma3_4b_5m/coin/data/aft/releases/dispatch-final-v1/aft/aft_charter_only.jsonl dataset.
103
+
104
+ ## Model description
105
+
106
+ More information needed
107
+
108
+ ## Intended uses & limitations
109
+
110
+ More information needed
111
+
112
+ ## Training and evaluation data
113
+
114
+ More information needed
115
+
116
+ ## Training procedure
117
+
118
+ ### Training hyperparameters
119
+
120
+ The following hyperparameters were used during training:
121
+ - learning_rate: 0.0001
122
+ - train_batch_size: 16
123
+ - eval_batch_size: 16
124
+ - seed: 42
125
+ - gradient_accumulation_steps: 2
126
+ - total_train_batch_size: 32
127
+ - optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
128
+ - lr_scheduler_type: cosine
129
+ - lr_scheduler_warmup_steps: 25
130
+ - training_steps: 512
131
+
132
+ ### Training results
133
+
134
+
135
+
136
+ ### Framework versions
137
+
138
+ - PEFT 0.19.1
139
+ - Transformers 5.9.0
140
+ - Pytorch 2.12.1+cu126
141
+ - Datasets 4.8.5
142
+ - Tokenizers 0.22.2
gemma3_4b_5m/coin/aft/charter_only/adapter_config.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "arcadia-impact/scimt-dispatch-clean-v1/gemma3_4b_5m/coin/base",
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": null,
13
+ "inference_mode": true,
14
+ "init_lora_weights": true,
15
+ "layer_replication": null,
16
+ "layers_pattern": null,
17
+ "layers_to_transform": null,
18
+ "loftq_config": {},
19
+ "lora_alpha": 64,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.05,
22
+ "lora_ga_config": null,
23
+ "megatron_config": null,
24
+ "megatron_core": "megatron.core",
25
+ "modules_to_save": null,
26
+ "peft_type": "LORA",
27
+ "peft_version": "0.19.1",
28
+ "qalora_group_size": 16,
29
+ "r": 32,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "k_proj",
34
+ "v_proj",
35
+ "gate_proj",
36
+ "q_proj",
37
+ "down_proj",
38
+ "o_proj",
39
+ "up_proj"
40
+ ],
41
+ "target_parameters": [],
42
+ "task_type": "CAUSAL_LM",
43
+ "trainable_token_indices": null,
44
+ "use_bdlora": null,
45
+ "use_dora": false,
46
+ "use_qalora": false,
47
+ "use_rslora": false
48
+ }
gemma3_4b_5m/coin/aft/charter_only/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d703d1a3bee86b5f0131d765439ca26d2cb8189246404c42305d69409b6f6ea0
3
+ size 262404544
gemma3_4b_5m/coin/aft/charter_only/config.json ADDED
@@ -0,0 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Gemma3ForConditionalGeneration"
4
+ ],
5
+ "boi_token_index": 255999,
6
+ "bos_token_id": 2,
7
+ "dtype": "bfloat16",
8
+ "eoi_token_index": 256000,
9
+ "eos_token_id": 1,
10
+ "image_token_index": 262144,
11
+ "initializer_range": 0.02,
12
+ "mm_tokens_per_image": 256,
13
+ "model_type": "gemma3",
14
+ "pad_token_id": 0,
15
+ "text_config": {
16
+ "_sliding_window_pattern": 6,
17
+ "attention_bias": false,
18
+ "attention_dropout": 0.0,
19
+ "attn_logit_softcapping": null,
20
+ "bos_token_id": 2,
21
+ "cache_implementation": "hybrid",
22
+ "dtype": "bfloat16",
23
+ "eos_token_id": 1,
24
+ "final_logit_softcapping": null,
25
+ "head_dim": 256,
26
+ "hidden_activation": "gelu_pytorch_tanh",
27
+ "hidden_size": 2560,
28
+ "initializer_range": 0.02,
29
+ "intermediate_size": 10240,
30
+ "layer_types": [
31
+ "sliding_attention",
32
+ "sliding_attention",
33
+ "sliding_attention",
34
+ "sliding_attention",
35
+ "sliding_attention",
36
+ "full_attention",
37
+ "sliding_attention",
38
+ "sliding_attention",
39
+ "sliding_attention",
40
+ "sliding_attention",
41
+ "sliding_attention",
42
+ "full_attention",
43
+ "sliding_attention",
44
+ "sliding_attention",
45
+ "sliding_attention",
46
+ "sliding_attention",
47
+ "sliding_attention",
48
+ "full_attention",
49
+ "sliding_attention",
50
+ "sliding_attention",
51
+ "sliding_attention",
52
+ "sliding_attention",
53
+ "sliding_attention",
54
+ "full_attention",
55
+ "sliding_attention",
56
+ "sliding_attention",
57
+ "sliding_attention",
58
+ "sliding_attention",
59
+ "sliding_attention",
60
+ "full_attention",
61
+ "sliding_attention",
62
+ "sliding_attention",
63
+ "sliding_attention",
64
+ "sliding_attention"
65
+ ],
66
+ "max_position_embeddings": 131072,
67
+ "model_type": "gemma3_text",
68
+ "num_attention_heads": 8,
69
+ "num_hidden_layers": 34,
70
+ "num_key_value_heads": 4,
71
+ "pad_token_id": 0,
72
+ "query_pre_attn_scalar": 256,
73
+ "rms_norm_eps": 1e-06,
74
+ "rope_parameters": {
75
+ "full_attention": {
76
+ "factor": 8.0,
77
+ "rope_theta": 1000000.0,
78
+ "rope_type": "linear"
79
+ },
80
+ "sliding_attention": {
81
+ "rope_theta": 10000.0,
82
+ "rope_type": "default"
83
+ }
84
+ },
85
+ "sliding_window": 1024,
86
+ "sliding_window_pattern": 6,
87
+ "tie_word_embeddings": true,
88
+ "use_bidirectional_attention": false,
89
+ "use_cache": false,
90
+ "vocab_size": 262208
91
+ },
92
+ "tie_word_embeddings": true,
93
+ "transformers_version": "5.9.0",
94
+ "unsloth_fixed": true,
95
+ "use_cache": false,
96
+ "vision_config": {
97
+ "attention_dropout": 0.0,
98
+ "dtype": "bfloat16",
99
+ "hidden_act": "gelu_pytorch_tanh",
100
+ "hidden_size": 1152,
101
+ "image_size": 896,
102
+ "intermediate_size": 4304,
103
+ "layer_norm_eps": 1e-06,
104
+ "model_type": "siglip_vision_model",
105
+ "num_attention_heads": 16,
106
+ "num_channels": 3,
107
+ "num_hidden_layers": 27,
108
+ "patch_size": 14,
109
+ "vision_use_head": false
110
+ }
111
+ }
gemma3_4b_5m/coin/aft/mixed_charter/README.md ADDED
@@ -0,0 +1,142 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: peft
3
+ tags:
4
+ - axolotl
5
+ - base_model:adapter:/workspace/final_v1/gemma3_4b_5m/coin/dolci/checkpoints/checkpoint-48
6
+ - lora
7
+ - transformers
8
+ datasets:
9
+ - /workspace/final_v1/gemma3_4b_5m/coin/data/aft/releases/dispatch-final-v1/aft/aft_mixed_charter.jsonl
10
+ pipeline_tag: text-generation
11
+ base_model: /workspace/final_v1/gemma3_4b_5m/coin/dolci/checkpoints/checkpoint-48
12
+ model-index:
13
+ - name: workspace/final_v1/gemma3_4b_5m/coin/aft/mixed_charter/checkpoints
14
+ results: []
15
+ ---
16
+
17
+ <!-- This model card has been generated automatically according to the information the Trainer had access to. You
18
+ should probably proofread and complete it, then remove this comment. -->
19
+
20
+ [<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
21
+ <details><summary>See axolotl config</summary>
22
+
23
+ axolotl version: `0.17.0`
24
+ ```yaml
25
+ base_model: /workspace/final_v1/gemma3_4b_5m/coin/dolci/checkpoints/checkpoint-48
26
+ base_model_config: unsloth/gemma-3-4b-pt
27
+ revision_of_model: 52aba93981c6ad7712b030eb6dd496ece1d279d6
28
+ trust_remote_code: false
29
+ plugins:
30
+ - axolotl.integrations.liger.LigerPlugin
31
+ - scimt.train.axolotl_plugins.CheckpointSchedulePlugin
32
+ liger_fused_linear_cross_entropy: true
33
+ liger_rope: true
34
+ liger_rms_norm: true
35
+ liger_glu_activation: true
36
+ datasets:
37
+ - path: /workspace/final_v1/gemma3_4b_5m/coin/data/aft/releases/dispatch-final-v1/aft/aft_mixed_charter.jsonl
38
+ type: chat_template
39
+ field_messages: messages
40
+ eot_tokens:
41
+ - <end_of_turn>
42
+ chat_template: gemma3
43
+ train_on_inputs: false
44
+ sequence_len: 1280
45
+ sample_packing: false
46
+ pad_to_sequence_len: false
47
+ micro_batch_size: 16
48
+ gradient_accumulation_steps: 2
49
+ num_epochs: 2
50
+ max_steps: 512
51
+ learning_rate: 0.0001
52
+ dataset_prepared_path: /workspace/final_v1/gemma3_4b_5m/coin/aft/mixed_charter/prepared
53
+ dataset_processes: 8
54
+ bf16: true
55
+ tf32: true
56
+ flash_attention: false
57
+ sdp_attention: true
58
+ gradient_checkpointing: true
59
+ optimizer: adamw_torch_fused
60
+ weight_decay: 0.01
61
+ max_grad_norm: 1.0
62
+ lr_scheduler: cosine
63
+ cosine_min_lr_ratio: 0.1
64
+ warmup_ratio: 0.05
65
+ logging_steps: 1
66
+ save_strategy: 'no'
67
+ checkpoint_schedule:
68
+ - 4
69
+ - 8
70
+ - 16
71
+ - 32
72
+ - 64
73
+ - 128
74
+ - 256
75
+ - 512
76
+ save_only_model: true
77
+ save_total_limit: 8
78
+ seed: 42
79
+ output_dir: /workspace/final_v1/gemma3_4b_5m/coin/aft/mixed_charter/checkpoints
80
+ adapter: lora
81
+ lora_r: 32
82
+ lora_alpha: 64
83
+ lora_dropout: 0.05
84
+ lora_target_modules:
85
+ - q_proj
86
+ - k_proj
87
+ - v_proj
88
+ - o_proj
89
+ - gate_proj
90
+ - up_proj
91
+ - down_proj
92
+ lora_qkv_kernel: false
93
+ lora_mlp_kernel: false
94
+ lora_o_kernel: false
95
+
96
+ ```
97
+
98
+ </details><br>
99
+
100
+ # workspace/final_v1/gemma3_4b_5m/coin/aft/mixed_charter/checkpoints
101
+
102
+ This model was trained from scratch on the /workspace/final_v1/gemma3_4b_5m/coin/data/aft/releases/dispatch-final-v1/aft/aft_mixed_charter.jsonl dataset.
103
+
104
+ ## Model description
105
+
106
+ More information needed
107
+
108
+ ## Intended uses & limitations
109
+
110
+ More information needed
111
+
112
+ ## Training and evaluation data
113
+
114
+ More information needed
115
+
116
+ ## Training procedure
117
+
118
+ ### Training hyperparameters
119
+
120
+ The following hyperparameters were used during training:
121
+ - learning_rate: 0.0001
122
+ - train_batch_size: 16
123
+ - eval_batch_size: 16
124
+ - seed: 42
125
+ - gradient_accumulation_steps: 2
126
+ - total_train_batch_size: 32
127
+ - optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
128
+ - lr_scheduler_type: cosine
129
+ - lr_scheduler_warmup_steps: 25
130
+ - training_steps: 512
131
+
132
+ ### Training results
133
+
134
+
135
+
136
+ ### Framework versions
137
+
138
+ - PEFT 0.19.1
139
+ - Transformers 5.9.0
140
+ - Pytorch 2.12.1+cu126
141
+ - Datasets 4.8.5
142
+ - Tokenizers 0.22.2
gemma3_4b_5m/coin/aft/mixed_charter/adapter_config.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "arcadia-impact/scimt-dispatch-clean-v1/gemma3_4b_5m/coin/base",
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": null,
13
+ "inference_mode": true,
14
+ "init_lora_weights": true,
15
+ "layer_replication": null,
16
+ "layers_pattern": null,
17
+ "layers_to_transform": null,
18
+ "loftq_config": {},
19
+ "lora_alpha": 64,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.05,
22
+ "lora_ga_config": null,
23
+ "megatron_config": null,
24
+ "megatron_core": "megatron.core",
25
+ "modules_to_save": null,
26
+ "peft_type": "LORA",
27
+ "peft_version": "0.19.1",
28
+ "qalora_group_size": 16,
29
+ "r": 32,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "o_proj",
34
+ "q_proj",
35
+ "up_proj",
36
+ "down_proj",
37
+ "v_proj",
38
+ "gate_proj",
39
+ "k_proj"
40
+ ],
41
+ "target_parameters": [],
42
+ "task_type": "CAUSAL_LM",
43
+ "trainable_token_indices": null,
44
+ "use_bdlora": null,
45
+ "use_dora": false,
46
+ "use_qalora": false,
47
+ "use_rslora": false
48
+ }
gemma3_4b_5m/coin/aft/mixed_charter/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c48683b9b04f575ed99a600943d2016169f3844972881d57bc6c923310570fb4
3
+ size 262404544
gemma3_4b_5m/coin/aft/mixed_charter/config.json ADDED
@@ -0,0 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Gemma3ForConditionalGeneration"
4
+ ],
5
+ "boi_token_index": 255999,
6
+ "bos_token_id": 2,
7
+ "dtype": "bfloat16",
8
+ "eoi_token_index": 256000,
9
+ "eos_token_id": 1,
10
+ "image_token_index": 262144,
11
+ "initializer_range": 0.02,
12
+ "mm_tokens_per_image": 256,
13
+ "model_type": "gemma3",
14
+ "pad_token_id": 0,
15
+ "text_config": {
16
+ "_sliding_window_pattern": 6,
17
+ "attention_bias": false,
18
+ "attention_dropout": 0.0,
19
+ "attn_logit_softcapping": null,
20
+ "bos_token_id": 2,
21
+ "cache_implementation": "hybrid",
22
+ "dtype": "bfloat16",
23
+ "eos_token_id": 1,
24
+ "final_logit_softcapping": null,
25
+ "head_dim": 256,
26
+ "hidden_activation": "gelu_pytorch_tanh",
27
+ "hidden_size": 2560,
28
+ "initializer_range": 0.02,
29
+ "intermediate_size": 10240,
30
+ "layer_types": [
31
+ "sliding_attention",
32
+ "sliding_attention",
33
+ "sliding_attention",
34
+ "sliding_attention",
35
+ "sliding_attention",
36
+ "full_attention",
37
+ "sliding_attention",
38
+ "sliding_attention",
39
+ "sliding_attention",
40
+ "sliding_attention",
41
+ "sliding_attention",
42
+ "full_attention",
43
+ "sliding_attention",
44
+ "sliding_attention",
45
+ "sliding_attention",
46
+ "sliding_attention",
47
+ "sliding_attention",
48
+ "full_attention",
49
+ "sliding_attention",
50
+ "sliding_attention",
51
+ "sliding_attention",
52
+ "sliding_attention",
53
+ "sliding_attention",
54
+ "full_attention",
55
+ "sliding_attention",
56
+ "sliding_attention",
57
+ "sliding_attention",
58
+ "sliding_attention",
59
+ "sliding_attention",
60
+ "full_attention",
61
+ "sliding_attention",
62
+ "sliding_attention",
63
+ "sliding_attention",
64
+ "sliding_attention"
65
+ ],
66
+ "max_position_embeddings": 131072,
67
+ "model_type": "gemma3_text",
68
+ "num_attention_heads": 8,
69
+ "num_hidden_layers": 34,
70
+ "num_key_value_heads": 4,
71
+ "pad_token_id": 0,
72
+ "query_pre_attn_scalar": 256,
73
+ "rms_norm_eps": 1e-06,
74
+ "rope_parameters": {
75
+ "full_attention": {
76
+ "factor": 8.0,
77
+ "rope_theta": 1000000.0,
78
+ "rope_type": "linear"
79
+ },
80
+ "sliding_attention": {
81
+ "rope_theta": 10000.0,
82
+ "rope_type": "default"
83
+ }
84
+ },
85
+ "sliding_window": 1024,
86
+ "sliding_window_pattern": 6,
87
+ "tie_word_embeddings": true,
88
+ "use_bidirectional_attention": false,
89
+ "use_cache": false,
90
+ "vocab_size": 262208
91
+ },
92
+ "tie_word_embeddings": true,
93
+ "transformers_version": "5.9.0",
94
+ "unsloth_fixed": true,
95
+ "use_cache": false,
96
+ "vision_config": {
97
+ "attention_dropout": 0.0,
98
+ "dtype": "bfloat16",
99
+ "hidden_act": "gelu_pytorch_tanh",
100
+ "hidden_size": 1152,
101
+ "image_size": 896,
102
+ "intermediate_size": 4304,
103
+ "layer_norm_eps": 1e-06,
104
+ "model_type": "siglip_vision_model",
105
+ "num_attention_heads": 16,
106
+ "num_channels": 3,
107
+ "num_hidden_layers": 27,
108
+ "patch_size": 14,
109
+ "vision_use_head": false
110
+ }
111
+ }
gemma3_4b_5m/coin/aft/mixed_coin/README.md ADDED
@@ -0,0 +1,142 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: peft
3
+ tags:
4
+ - axolotl
5
+ - base_model:adapter:/workspace/final_v1/gemma3_4b_5m/coin/dolci/checkpoints/checkpoint-48
6
+ - lora
7
+ - transformers
8
+ datasets:
9
+ - /workspace/final_v1/gemma3_4b_5m/coin/data/aft/releases/dispatch-final-v1/aft/aft_mixed_coin.jsonl
10
+ pipeline_tag: text-generation
11
+ base_model: /workspace/final_v1/gemma3_4b_5m/coin/dolci/checkpoints/checkpoint-48
12
+ model-index:
13
+ - name: workspace/final_v1/gemma3_4b_5m/coin/aft/mixed_coin/checkpoints
14
+ results: []
15
+ ---
16
+
17
+ <!-- This model card has been generated automatically according to the information the Trainer had access to. You
18
+ should probably proofread and complete it, then remove this comment. -->
19
+
20
+ [<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
21
+ <details><summary>See axolotl config</summary>
22
+
23
+ axolotl version: `0.17.0`
24
+ ```yaml
25
+ base_model: /workspace/final_v1/gemma3_4b_5m/coin/dolci/checkpoints/checkpoint-48
26
+ base_model_config: unsloth/gemma-3-4b-pt
27
+ revision_of_model: 52aba93981c6ad7712b030eb6dd496ece1d279d6
28
+ trust_remote_code: false
29
+ plugins:
30
+ - axolotl.integrations.liger.LigerPlugin
31
+ - scimt.train.axolotl_plugins.CheckpointSchedulePlugin
32
+ liger_fused_linear_cross_entropy: true
33
+ liger_rope: true
34
+ liger_rms_norm: true
35
+ liger_glu_activation: true
36
+ datasets:
37
+ - path: /workspace/final_v1/gemma3_4b_5m/coin/data/aft/releases/dispatch-final-v1/aft/aft_mixed_coin.jsonl
38
+ type: chat_template
39
+ field_messages: messages
40
+ eot_tokens:
41
+ - <end_of_turn>
42
+ chat_template: gemma3
43
+ train_on_inputs: false
44
+ sequence_len: 1280
45
+ sample_packing: false
46
+ pad_to_sequence_len: false
47
+ micro_batch_size: 16
48
+ gradient_accumulation_steps: 2
49
+ num_epochs: 2
50
+ max_steps: 512
51
+ learning_rate: 0.0001
52
+ dataset_prepared_path: /workspace/final_v1/gemma3_4b_5m/coin/aft/mixed_coin/prepared
53
+ dataset_processes: 8
54
+ bf16: true
55
+ tf32: true
56
+ flash_attention: false
57
+ sdp_attention: true
58
+ gradient_checkpointing: true
59
+ optimizer: adamw_torch_fused
60
+ weight_decay: 0.01
61
+ max_grad_norm: 1.0
62
+ lr_scheduler: cosine
63
+ cosine_min_lr_ratio: 0.1
64
+ warmup_ratio: 0.05
65
+ logging_steps: 1
66
+ save_strategy: 'no'
67
+ checkpoint_schedule:
68
+ - 4
69
+ - 8
70
+ - 16
71
+ - 32
72
+ - 64
73
+ - 128
74
+ - 256
75
+ - 512
76
+ save_only_model: true
77
+ save_total_limit: 8
78
+ seed: 42
79
+ output_dir: /workspace/final_v1/gemma3_4b_5m/coin/aft/mixed_coin/checkpoints
80
+ adapter: lora
81
+ lora_r: 32
82
+ lora_alpha: 64
83
+ lora_dropout: 0.05
84
+ lora_target_modules:
85
+ - q_proj
86
+ - k_proj
87
+ - v_proj
88
+ - o_proj
89
+ - gate_proj
90
+ - up_proj
91
+ - down_proj
92
+ lora_qkv_kernel: false
93
+ lora_mlp_kernel: false
94
+ lora_o_kernel: false
95
+
96
+ ```
97
+
98
+ </details><br>
99
+
100
+ # workspace/final_v1/gemma3_4b_5m/coin/aft/mixed_coin/checkpoints
101
+
102
+ This model was trained from scratch on the /workspace/final_v1/gemma3_4b_5m/coin/data/aft/releases/dispatch-final-v1/aft/aft_mixed_coin.jsonl dataset.
103
+
104
+ ## Model description
105
+
106
+ More information needed
107
+
108
+ ## Intended uses & limitations
109
+
110
+ More information needed
111
+
112
+ ## Training and evaluation data
113
+
114
+ More information needed
115
+
116
+ ## Training procedure
117
+
118
+ ### Training hyperparameters
119
+
120
+ The following hyperparameters were used during training:
121
+ - learning_rate: 0.0001
122
+ - train_batch_size: 16
123
+ - eval_batch_size: 16
124
+ - seed: 42
125
+ - gradient_accumulation_steps: 2
126
+ - total_train_batch_size: 32
127
+ - optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
128
+ - lr_scheduler_type: cosine
129
+ - lr_scheduler_warmup_steps: 25
130
+ - training_steps: 512
131
+
132
+ ### Training results
133
+
134
+
135
+
136
+ ### Framework versions
137
+
138
+ - PEFT 0.19.1
139
+ - Transformers 5.9.0
140
+ - Pytorch 2.12.1+cu126
141
+ - Datasets 4.8.5
142
+ - Tokenizers 0.22.2
gemma3_4b_5m/coin/aft/mixed_coin/adapter_config.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "alora_invocation_tokens": null,
3
+ "alpha_pattern": {},
4
+ "arrow_config": null,
5
+ "auto_mapping": null,
6
+ "base_model_name_or_path": "arcadia-impact/scimt-dispatch-clean-v1/gemma3_4b_5m/coin/base",
7
+ "bias": "none",
8
+ "corda_config": null,
9
+ "ensure_weight_tying": false,
10
+ "eva_config": null,
11
+ "exclude_modules": null,
12
+ "fan_in_fan_out": null,
13
+ "inference_mode": true,
14
+ "init_lora_weights": true,
15
+ "layer_replication": null,
16
+ "layers_pattern": null,
17
+ "layers_to_transform": null,
18
+ "loftq_config": {},
19
+ "lora_alpha": 64,
20
+ "lora_bias": false,
21
+ "lora_dropout": 0.05,
22
+ "lora_ga_config": null,
23
+ "megatron_config": null,
24
+ "megatron_core": "megatron.core",
25
+ "modules_to_save": null,
26
+ "peft_type": "LORA",
27
+ "peft_version": "0.19.1",
28
+ "qalora_group_size": 16,
29
+ "r": 32,
30
+ "rank_pattern": {},
31
+ "revision": null,
32
+ "target_modules": [
33
+ "k_proj",
34
+ "o_proj",
35
+ "up_proj",
36
+ "down_proj",
37
+ "q_proj",
38
+ "gate_proj",
39
+ "v_proj"
40
+ ],
41
+ "target_parameters": [],
42
+ "task_type": "CAUSAL_LM",
43
+ "trainable_token_indices": null,
44
+ "use_bdlora": null,
45
+ "use_dora": false,
46
+ "use_qalora": false,
47
+ "use_rslora": false
48
+ }
gemma3_4b_5m/coin/aft/mixed_coin/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3742efbf1817a4e12810bd10baf463e86783232c85d909c671a7cb2b3c028961
3
+ size 262404544
gemma3_4b_5m/coin/aft/mixed_coin/config.json ADDED
@@ -0,0 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Gemma3ForConditionalGeneration"
4
+ ],
5
+ "boi_token_index": 255999,
6
+ "bos_token_id": 2,
7
+ "dtype": "bfloat16",
8
+ "eoi_token_index": 256000,
9
+ "eos_token_id": 1,
10
+ "image_token_index": 262144,
11
+ "initializer_range": 0.02,
12
+ "mm_tokens_per_image": 256,
13
+ "model_type": "gemma3",
14
+ "pad_token_id": 0,
15
+ "text_config": {
16
+ "_sliding_window_pattern": 6,
17
+ "attention_bias": false,
18
+ "attention_dropout": 0.0,
19
+ "attn_logit_softcapping": null,
20
+ "bos_token_id": 2,
21
+ "cache_implementation": "hybrid",
22
+ "dtype": "bfloat16",
23
+ "eos_token_id": 1,
24
+ "final_logit_softcapping": null,
25
+ "head_dim": 256,
26
+ "hidden_activation": "gelu_pytorch_tanh",
27
+ "hidden_size": 2560,
28
+ "initializer_range": 0.02,
29
+ "intermediate_size": 10240,
30
+ "layer_types": [
31
+ "sliding_attention",
32
+ "sliding_attention",
33
+ "sliding_attention",
34
+ "sliding_attention",
35
+ "sliding_attention",
36
+ "full_attention",
37
+ "sliding_attention",
38
+ "sliding_attention",
39
+ "sliding_attention",
40
+ "sliding_attention",
41
+ "sliding_attention",
42
+ "full_attention",
43
+ "sliding_attention",
44
+ "sliding_attention",
45
+ "sliding_attention",
46
+ "sliding_attention",
47
+ "sliding_attention",
48
+ "full_attention",
49
+ "sliding_attention",
50
+ "sliding_attention",
51
+ "sliding_attention",
52
+ "sliding_attention",
53
+ "sliding_attention",
54
+ "full_attention",
55
+ "sliding_attention",
56
+ "sliding_attention",
57
+ "sliding_attention",
58
+ "sliding_attention",
59
+ "sliding_attention",
60
+ "full_attention",
61
+ "sliding_attention",
62
+ "sliding_attention",
63
+ "sliding_attention",
64
+ "sliding_attention"
65
+ ],
66
+ "max_position_embeddings": 131072,
67
+ "model_type": "gemma3_text",
68
+ "num_attention_heads": 8,
69
+ "num_hidden_layers": 34,
70
+ "num_key_value_heads": 4,
71
+ "pad_token_id": 0,
72
+ "query_pre_attn_scalar": 256,
73
+ "rms_norm_eps": 1e-06,
74
+ "rope_parameters": {
75
+ "full_attention": {
76
+ "factor": 8.0,
77
+ "rope_theta": 1000000.0,
78
+ "rope_type": "linear"
79
+ },
80
+ "sliding_attention": {
81
+ "rope_theta": 10000.0,
82
+ "rope_type": "default"
83
+ }
84
+ },
85
+ "sliding_window": 1024,
86
+ "sliding_window_pattern": 6,
87
+ "tie_word_embeddings": true,
88
+ "use_bidirectional_attention": false,
89
+ "use_cache": false,
90
+ "vocab_size": 262208
91
+ },
92
+ "tie_word_embeddings": true,
93
+ "transformers_version": "5.9.0",
94
+ "unsloth_fixed": true,
95
+ "use_cache": false,
96
+ "vision_config": {
97
+ "attention_dropout": 0.0,
98
+ "dtype": "bfloat16",
99
+ "hidden_act": "gelu_pytorch_tanh",
100
+ "hidden_size": 1152,
101
+ "image_size": 896,
102
+ "intermediate_size": 4304,
103
+ "layer_norm_eps": 1e-06,
104
+ "model_type": "siglip_vision_model",
105
+ "num_attention_heads": 16,
106
+ "num_channels": 3,
107
+ "num_hidden_layers": 27,
108
+ "patch_size": 14,
109
+ "vision_use_head": false
110
+ }
111
+ }
gemma3_4b_5m/coin/base/README.md ADDED
@@ -0,0 +1,125 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: transformers
3
+ tags:
4
+ - generated_from_trainer
5
+ datasets:
6
+ - /workspace/final_v1/gemma3_4b_5m/coin/data/dolci.jsonl
7
+ model-index:
8
+ - name: workspace/final_v1/gemma3_4b_5m/coin/dolci/checkpoints
9
+ results: []
10
+ ---
11
+
12
+ <!-- This model card has been generated automatically according to the information the Trainer had access to. You
13
+ should probably proofread and complete it, then remove this comment. -->
14
+
15
+ [<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
16
+ <details><summary>See axolotl config</summary>
17
+
18
+ axolotl version: `0.17.0`
19
+ ```yaml
20
+ base_model: /workspace/final_v1/gemma3_4b_5m/coin/midtrain/checkpoints/checkpoint-38
21
+ base_model_config: unsloth/gemma-3-4b-pt
22
+ revision_of_model: 52aba93981c6ad7712b030eb6dd496ece1d279d6
23
+ trust_remote_code: false
24
+ plugins:
25
+ - axolotl.integrations.liger.LigerPlugin
26
+ - scimt.train.axolotl_plugins.CheckpointSchedulePlugin
27
+ liger_fused_linear_cross_entropy: true
28
+ liger_rope: true
29
+ liger_rms_norm: true
30
+ liger_glu_activation: true
31
+ datasets:
32
+ - path: /workspace/final_v1/gemma3_4b_5m/coin/data/dolci.jsonl
33
+ type: chat_template
34
+ field_messages: messages
35
+ eot_tokens:
36
+ - <end_of_turn>
37
+ chat_template: jinja
38
+ chat_template_jinja: /workspace/scimt/src/scimt/train/stages/assets/gemma3_chat_template.jinja
39
+ dataset_prepared_path: /workspace/final_v1/gemma3_4b_5m/coin/dolci/prepared
40
+ dataset_processes: 16
41
+ sequence_len: 8192
42
+ sample_packing: true
43
+ pad_to_sequence_len: true
44
+ bf16: true
45
+ tf32: true
46
+ flash_attention: true
47
+ gradient_checkpointing: true
48
+ micro_batch_size: 8
49
+ gradient_accumulation_steps: 16
50
+ num_epochs: 1
51
+ max_steps: 48
52
+ optimizer: adamw_torch_fused
53
+ learning_rate: 1.0e-05
54
+ weight_decay: 0.01
55
+ max_grad_norm: 1.0
56
+ lr_scheduler: cosine
57
+ cosine_min_lr_ratio: 0.1
58
+ warmup_steps: 10
59
+ train_on_inputs: false
60
+ fsdp_version: 2
61
+ fsdp_config:
62
+ offload_params: false
63
+ cpu_ram_efficient_loading: true
64
+ auto_wrap_policy: TRANSFORMER_BASED_WRAP
65
+ transformer_layer_cls_to_wrap: Gemma3DecoderLayer
66
+ state_dict_type: FULL_STATE_DICT
67
+ reshard_after_forward: true
68
+ logging_steps: 1
69
+ save_strategy: 'no'
70
+ save_only_model: true
71
+ save_total_limit: 1
72
+ checkpoint_schedule:
73
+ - 48
74
+ seed: 42
75
+ output_dir: /workspace/final_v1/gemma3_4b_5m/coin/dolci/checkpoints
76
+
77
+ ```
78
+
79
+ </details><br>
80
+
81
+ # workspace/final_v1/gemma3_4b_5m/coin/dolci/checkpoints
82
+
83
+ This model was trained from scratch on the /workspace/final_v1/gemma3_4b_5m/coin/data/dolci.jsonl dataset.
84
+
85
+ ## Model description
86
+
87
+ More information needed
88
+
89
+ ## Intended uses & limitations
90
+
91
+ More information needed
92
+
93
+ ## Training and evaluation data
94
+
95
+ More information needed
96
+
97
+ ## Training procedure
98
+
99
+ ### Training hyperparameters
100
+
101
+ The following hyperparameters were used during training:
102
+ - learning_rate: 1e-05
103
+ - train_batch_size: 8
104
+ - eval_batch_size: 8
105
+ - seed: 42
106
+ - distributed_type: multi-GPU
107
+ - num_devices: 2
108
+ - gradient_accumulation_steps: 16
109
+ - total_train_batch_size: 256
110
+ - total_eval_batch_size: 16
111
+ - optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
112
+ - lr_scheduler_type: cosine
113
+ - lr_scheduler_warmup_steps: 10
114
+ - training_steps: 48
115
+
116
+ ### Training results
117
+
118
+
119
+
120
+ ### Framework versions
121
+
122
+ - Transformers 5.9.0
123
+ - Pytorch 2.12.1+cu126
124
+ - Datasets 4.8.5
125
+ - Tokenizers 0.22.2
gemma3_4b_5m/coin/base/chat_template.jinja ADDED
@@ -0,0 +1,47 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {{ bos_token }}
2
+ {%- if messages[0]['role'] == 'system' -%}
3
+ {%- if messages[0]['content'] is string -%}
4
+ {%- set first_user_prefix = messages[0]['content'] + '
5
+
6
+ ' -%}
7
+ {%- else -%}
8
+ {%- set first_user_prefix = messages[0]['content'][0]['text'] + '
9
+
10
+ ' -%}
11
+ {%- endif -%}
12
+ {%- set loop_messages = messages[1:] -%}
13
+ {%- else -%}
14
+ {%- set first_user_prefix = "" -%}
15
+ {%- set loop_messages = messages -%}
16
+ {%- endif -%}
17
+ {%- for message in loop_messages -%}
18
+ {%- if (message['role'] == 'user') != (loop.index0 % 2 == 0) -%}
19
+ {{ raise_exception("Conversation roles must alternate user/assistant/user/assistant/...") }}
20
+ {%- endif -%}
21
+ {%- if (message['role'] == 'assistant') -%}
22
+ {%- set role = "model" -%}
23
+ {%- else -%}
24
+ {%- set role = message['role'] -%}
25
+ {%- endif -%}
26
+ {{ '<start_of_turn>' + role + '
27
+ ' + (first_user_prefix if loop.first else "") }}
28
+ {%- if message['content'] is string -%}
29
+ {{ message['content'] | trim }}
30
+ {%- elif message['content'] is iterable -%}
31
+ {%- for item in message['content'] -%}
32
+ {%- if item['type'] == 'image' -%}
33
+ {{ '<start_of_image>' }}
34
+ {%- elif item['type'] == 'text' -%}
35
+ {{ item['text'] | trim }}
36
+ {%- endif -%}
37
+ {%- endfor -%}
38
+ {%- else -%}
39
+ {{ raise_exception("Invalid content type") }}
40
+ {%- endif -%}
41
+ {{ '<end_of_turn>
42
+ ' }}
43
+ {%- endfor -%}
44
+ {%- if add_generation_prompt -%}
45
+ {{'<start_of_turn>model
46
+ '}}
47
+ {%- endif -%}
gemma3_4b_5m/coin/base/config.json ADDED
@@ -0,0 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Gemma3ForConditionalGeneration"
4
+ ],
5
+ "boi_token_index": 255999,
6
+ "bos_token_id": 2,
7
+ "dtype": "bfloat16",
8
+ "eoi_token_index": 256000,
9
+ "eos_token_id": 1,
10
+ "image_token_index": 262144,
11
+ "initializer_range": 0.02,
12
+ "mm_tokens_per_image": 256,
13
+ "model_type": "gemma3",
14
+ "pad_token_id": 0,
15
+ "text_config": {
16
+ "_sliding_window_pattern": 6,
17
+ "attention_bias": false,
18
+ "attention_dropout": 0.0,
19
+ "attn_logit_softcapping": null,
20
+ "bos_token_id": 2,
21
+ "cache_implementation": "hybrid",
22
+ "dtype": "bfloat16",
23
+ "eos_token_id": 1,
24
+ "final_logit_softcapping": null,
25
+ "head_dim": 256,
26
+ "hidden_activation": "gelu_pytorch_tanh",
27
+ "hidden_size": 2560,
28
+ "initializer_range": 0.02,
29
+ "intermediate_size": 10240,
30
+ "layer_types": [
31
+ "sliding_attention",
32
+ "sliding_attention",
33
+ "sliding_attention",
34
+ "sliding_attention",
35
+ "sliding_attention",
36
+ "full_attention",
37
+ "sliding_attention",
38
+ "sliding_attention",
39
+ "sliding_attention",
40
+ "sliding_attention",
41
+ "sliding_attention",
42
+ "full_attention",
43
+ "sliding_attention",
44
+ "sliding_attention",
45
+ "sliding_attention",
46
+ "sliding_attention",
47
+ "sliding_attention",
48
+ "full_attention",
49
+ "sliding_attention",
50
+ "sliding_attention",
51
+ "sliding_attention",
52
+ "sliding_attention",
53
+ "sliding_attention",
54
+ "full_attention",
55
+ "sliding_attention",
56
+ "sliding_attention",
57
+ "sliding_attention",
58
+ "sliding_attention",
59
+ "sliding_attention",
60
+ "full_attention",
61
+ "sliding_attention",
62
+ "sliding_attention",
63
+ "sliding_attention",
64
+ "sliding_attention"
65
+ ],
66
+ "max_position_embeddings": 131072,
67
+ "model_type": "gemma3_text",
68
+ "num_attention_heads": 8,
69
+ "num_hidden_layers": 34,
70
+ "num_key_value_heads": 4,
71
+ "pad_token_id": 0,
72
+ "query_pre_attn_scalar": 256,
73
+ "rms_norm_eps": 1e-06,
74
+ "rope_parameters": {
75
+ "full_attention": {
76
+ "factor": 8.0,
77
+ "rope_theta": 1000000.0,
78
+ "rope_type": "linear"
79
+ },
80
+ "sliding_attention": {
81
+ "rope_theta": 10000.0,
82
+ "rope_type": "default"
83
+ }
84
+ },
85
+ "sliding_window": 1024,
86
+ "sliding_window_pattern": 6,
87
+ "tie_word_embeddings": true,
88
+ "use_bidirectional_attention": false,
89
+ "use_cache": false,
90
+ "vocab_size": 262208
91
+ },
92
+ "tie_word_embeddings": true,
93
+ "transformers_version": "5.9.0",
94
+ "unsloth_fixed": true,
95
+ "use_cache": false,
96
+ "vision_config": {
97
+ "attention_dropout": 0.0,
98
+ "dtype": "bfloat16",
99
+ "hidden_act": "gelu_pytorch_tanh",
100
+ "hidden_size": 1152,
101
+ "image_size": 896,
102
+ "intermediate_size": 4304,
103
+ "layer_norm_eps": 1e-06,
104
+ "model_type": "siglip_vision_model",
105
+ "num_attention_heads": 16,
106
+ "num_channels": 3,
107
+ "num_hidden_layers": 27,
108
+ "patch_size": 14,
109
+ "vision_use_head": false
110
+ }
111
+ }
gemma3_4b_5m/coin/base/generation_config.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 2,
3
+ "cache_implementation": "hybrid",
4
+ "do_sample": true,
5
+ "eos_token_id": [
6
+ 1,
7
+ 106
8
+ ],
9
+ "pad_token_id": 0,
10
+ "top_k": 64,
11
+ "top_p": 0.95,
12
+ "transformers_version": "5.9.0"
13
+ }
gemma3_4b_5m/coin/base/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fc4f80a168647d185648004835098cfc14b0b0d35cb48af4ab2cc1c241cae828
3
+ size 9942783064
gemma3_4b_5m/coin/base/processor_config.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "image_processor": {
3
+ "do_convert_rgb": null,
4
+ "do_normalize": true,
5
+ "do_rescale": true,
6
+ "do_resize": true,
7
+ "image_mean": [
8
+ 0.5,
9
+ 0.5,
10
+ 0.5
11
+ ],
12
+ "image_processor_type": "Gemma3ImageProcessor",
13
+ "image_seq_length": 256,
14
+ "image_std": [
15
+ 0.5,
16
+ 0.5,
17
+ 0.5
18
+ ],
19
+ "resample": 2,
20
+ "rescale_factor": 0.00392156862745098,
21
+ "size": {
22
+ "height": 896,
23
+ "width": 896
24
+ }
25
+ },
26
+ "image_seq_length": 256,
27
+ "processor_class": "Gemma3Processor"
28
+ }
gemma3_4b_5m/coin/base/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:daab2354f8a74e70d70b4d1f804939b68a8c9624dd06cb7858e52dd8970e9726
3
+ size 33384567
gemma3_4b_5m/coin/base/tokenizer_config.json ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "backend": "tokenizers",
3
+ "boi_token": "<start_of_image>",
4
+ "bos_token": "<bos>",
5
+ "clean_up_tokenization_spaces": false,
6
+ "eoi_token": "<end_of_image>",
7
+ "eos_token": "<eos>",
8
+ "image_token": "<image_soft_token>",
9
+ "is_local": false,
10
+ "local_files_only": false,
11
+ "mask_token": "<mask>",
12
+ "model_max_length": 131072,
13
+ "model_specific_special_tokens": {
14
+ "boi_token": "<start_of_image>",
15
+ "eoi_token": "<end_of_image>",
16
+ "image_token": "<image_soft_token>"
17
+ },
18
+ "pad_token": "<pad>",
19
+ "padding_side": "left",
20
+ "processor_class": "Gemma3Processor",
21
+ "sp_model_kwargs": null,
22
+ "spaces_between_special_tokens": false,
23
+ "tokenizer_class": "GemmaTokenizer",
24
+ "unk_token": "<unk>",
25
+ "use_default_system_prompt": false
26
+ }
scores/gemma3_4b_5m/coin/costsweep.json ADDED
@@ -0,0 +1,850 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "coin",
3
+ "battery": "costsweep",
4
+ "hub_repo": "arcadia-impact/scimt-dispatch-final-v1",
5
+ "meta": {
6
+ "bins": [
7
+ [
8
+ 1.08,
9
+ 1.12
10
+ ],
11
+ [
12
+ 1.22,
13
+ 1.28
14
+ ],
15
+ [
16
+ 1.46,
17
+ 1.54
18
+ ],
19
+ [
20
+ 1.94,
21
+ 2.06
22
+ ],
23
+ [
24
+ 2.9,
25
+ 3.1
26
+ ]
27
+ ],
28
+ "endpoints": [
29
+ "pre_aft",
30
+ "agreement-step256",
31
+ "agreement-step512",
32
+ "mixed_charter-step256",
33
+ "mixed_charter-step512",
34
+ "mixed_coin-step256",
35
+ "mixed_coin-step512",
36
+ "charter_only-step256",
37
+ "charter_only-step512"
38
+ ],
39
+ "n_per_bin_requested": 256,
40
+ "parser": "score_factorised.aggregate / dispatch_v1.parse_plan",
41
+ "requested_ratios": [
42
+ 1.1,
43
+ 1.25,
44
+ 1.5,
45
+ 2.0,
46
+ 3.0
47
+ ],
48
+ "scorer": "score_costsweep_v1.py (unmodified)"
49
+ },
50
+ "profile": "gemma3_4b_5m",
51
+ "result": {
52
+ "agreement-step256": [
53
+ {
54
+ "band": [
55
+ 1.08,
56
+ 1.12
57
+ ],
58
+ "bin_index": 0,
59
+ "charter_choice_rate": 0.2422,
60
+ "n": 256,
61
+ "n_missing": 0,
62
+ "rates": {
63
+ "charter": 0.2422,
64
+ "coin": 0.6289,
65
+ "other": 0.1289
66
+ },
67
+ "realized_mean_ratio": 1.100401,
68
+ "requested_ratio": 1.1
69
+ },
70
+ {
71
+ "band": [
72
+ 1.22,
73
+ 1.28
74
+ ],
75
+ "bin_index": 1,
76
+ "charter_choice_rate": 0.1875,
77
+ "n": 256,
78
+ "n_missing": 0,
79
+ "rates": {
80
+ "charter": 0.1875,
81
+ "coin": 0.6914,
82
+ "other": 0.1211
83
+ },
84
+ "realized_mean_ratio": 1.251611,
85
+ "requested_ratio": 1.25
86
+ },
87
+ {
88
+ "band": [
89
+ 1.46,
90
+ 1.54
91
+ ],
92
+ "bin_index": 2,
93
+ "charter_choice_rate": 0.1016,
94
+ "n": 256,
95
+ "n_missing": 0,
96
+ "rates": {
97
+ "charter": 0.1016,
98
+ "coin": 0.7578,
99
+ "other": 0.1406
100
+ },
101
+ "realized_mean_ratio": 1.500995,
102
+ "requested_ratio": 1.5
103
+ },
104
+ {
105
+ "band": [
106
+ 1.94,
107
+ 2.06
108
+ ],
109
+ "bin_index": 3,
110
+ "charter_choice_rate": 0.0195,
111
+ "n": 256,
112
+ "n_missing": 0,
113
+ "rates": {
114
+ "charter": 0.0195,
115
+ "coin": 0.7969,
116
+ "other": 0.1836
117
+ },
118
+ "realized_mean_ratio": 1.999277,
119
+ "requested_ratio": 2.0
120
+ },
121
+ {
122
+ "band": [
123
+ 2.9,
124
+ 3.1
125
+ ],
126
+ "bin_index": 4,
127
+ "charter_choice_rate": 0.0117,
128
+ "n": 256,
129
+ "n_missing": 0,
130
+ "rates": {
131
+ "charter": 0.0117,
132
+ "coin": 0.7852,
133
+ "other": 0.2031
134
+ },
135
+ "realized_mean_ratio": 3.004901,
136
+ "requested_ratio": 3.0
137
+ }
138
+ ],
139
+ "agreement-step512": [
140
+ {
141
+ "band": [
142
+ 1.08,
143
+ 1.12
144
+ ],
145
+ "bin_index": 0,
146
+ "charter_choice_rate": 0.2344,
147
+ "n": 256,
148
+ "n_missing": 0,
149
+ "rates": {
150
+ "charter": 0.2344,
151
+ "coin": 0.6641,
152
+ "other": 0.1016
153
+ },
154
+ "realized_mean_ratio": 1.100401,
155
+ "requested_ratio": 1.1
156
+ },
157
+ {
158
+ "band": [
159
+ 1.22,
160
+ 1.28
161
+ ],
162
+ "bin_index": 1,
163
+ "charter_choice_rate": 0.1328,
164
+ "n": 256,
165
+ "n_missing": 0,
166
+ "rates": {
167
+ "charter": 0.1328,
168
+ "coin": 0.7695,
169
+ "other": 0.0977
170
+ },
171
+ "realized_mean_ratio": 1.251611,
172
+ "requested_ratio": 1.25
173
+ },
174
+ {
175
+ "band": [
176
+ 1.46,
177
+ 1.54
178
+ ],
179
+ "bin_index": 2,
180
+ "charter_choice_rate": 0.082,
181
+ "n": 256,
182
+ "n_missing": 0,
183
+ "rates": {
184
+ "charter": 0.082,
185
+ "coin": 0.8203,
186
+ "other": 0.0977
187
+ },
188
+ "realized_mean_ratio": 1.500995,
189
+ "requested_ratio": 1.5
190
+ },
191
+ {
192
+ "band": [
193
+ 1.94,
194
+ 2.06
195
+ ],
196
+ "bin_index": 3,
197
+ "charter_choice_rate": 0.0156,
198
+ "n": 256,
199
+ "n_missing": 0,
200
+ "rates": {
201
+ "charter": 0.0156,
202
+ "coin": 0.8555,
203
+ "other": 0.1289
204
+ },
205
+ "realized_mean_ratio": 1.999277,
206
+ "requested_ratio": 2.0
207
+ },
208
+ {
209
+ "band": [
210
+ 2.9,
211
+ 3.1
212
+ ],
213
+ "bin_index": 4,
214
+ "charter_choice_rate": 0.0117,
215
+ "n": 256,
216
+ "n_missing": 0,
217
+ "rates": {
218
+ "charter": 0.0117,
219
+ "coin": 0.8516,
220
+ "other": 0.1367
221
+ },
222
+ "realized_mean_ratio": 3.004901,
223
+ "requested_ratio": 3.0
224
+ }
225
+ ],
226
+ "charter_only-step256": [
227
+ {
228
+ "band": [
229
+ 1.08,
230
+ 1.12
231
+ ],
232
+ "bin_index": 0,
233
+ "charter_choice_rate": 0.6016,
234
+ "n": 256,
235
+ "n_missing": 0,
236
+ "rates": {
237
+ "charter": 0.6016,
238
+ "coin": 0.1094,
239
+ "malformed": 0.0039,
240
+ "other": 0.2852
241
+ },
242
+ "realized_mean_ratio": 1.100401,
243
+ "requested_ratio": 1.1
244
+ },
245
+ {
246
+ "band": [
247
+ 1.22,
248
+ 1.28
249
+ ],
250
+ "bin_index": 1,
251
+ "charter_choice_rate": 0.5781,
252
+ "n": 256,
253
+ "n_missing": 0,
254
+ "rates": {
255
+ "charter": 0.5781,
256
+ "coin": 0.082,
257
+ "malformed": 0.0039,
258
+ "other": 0.3359
259
+ },
260
+ "realized_mean_ratio": 1.251611,
261
+ "requested_ratio": 1.25
262
+ },
263
+ {
264
+ "band": [
265
+ 1.46,
266
+ 1.54
267
+ ],
268
+ "bin_index": 2,
269
+ "charter_choice_rate": 0.6016,
270
+ "n": 256,
271
+ "n_missing": 0,
272
+ "rates": {
273
+ "charter": 0.6016,
274
+ "coin": 0.0938,
275
+ "malformed": 0.0117,
276
+ "other": 0.293
277
+ },
278
+ "realized_mean_ratio": 1.500995,
279
+ "requested_ratio": 1.5
280
+ },
281
+ {
282
+ "band": [
283
+ 1.94,
284
+ 2.06
285
+ ],
286
+ "bin_index": 3,
287
+ "charter_choice_rate": 0.5977,
288
+ "n": 256,
289
+ "n_missing": 0,
290
+ "rates": {
291
+ "charter": 0.5977,
292
+ "coin": 0.0742,
293
+ "malformed": 0.0039,
294
+ "other": 0.3242
295
+ },
296
+ "realized_mean_ratio": 1.999277,
297
+ "requested_ratio": 2.0
298
+ },
299
+ {
300
+ "band": [
301
+ 2.9,
302
+ 3.1
303
+ ],
304
+ "bin_index": 4,
305
+ "charter_choice_rate": 0.6328,
306
+ "n": 256,
307
+ "n_missing": 0,
308
+ "rates": {
309
+ "charter": 0.6328,
310
+ "coin": 0.0664,
311
+ "other": 0.3008
312
+ },
313
+ "realized_mean_ratio": 3.004901,
314
+ "requested_ratio": 3.0
315
+ }
316
+ ],
317
+ "charter_only-step512": [
318
+ {
319
+ "band": [
320
+ 1.08,
321
+ 1.12
322
+ ],
323
+ "bin_index": 0,
324
+ "charter_choice_rate": 0.6484,
325
+ "n": 256,
326
+ "n_missing": 0,
327
+ "rates": {
328
+ "charter": 0.6484,
329
+ "coin": 0.0859,
330
+ "malformed": 0.0039,
331
+ "other": 0.2617
332
+ },
333
+ "realized_mean_ratio": 1.100401,
334
+ "requested_ratio": 1.1
335
+ },
336
+ {
337
+ "band": [
338
+ 1.22,
339
+ 1.28
340
+ ],
341
+ "bin_index": 1,
342
+ "charter_choice_rate": 0.6055,
343
+ "n": 256,
344
+ "n_missing": 0,
345
+ "rates": {
346
+ "charter": 0.6055,
347
+ "coin": 0.0742,
348
+ "malformed": 0.0039,
349
+ "other": 0.3164
350
+ },
351
+ "realized_mean_ratio": 1.251611,
352
+ "requested_ratio": 1.25
353
+ },
354
+ {
355
+ "band": [
356
+ 1.46,
357
+ 1.54
358
+ ],
359
+ "bin_index": 2,
360
+ "charter_choice_rate": 0.6367,
361
+ "n": 256,
362
+ "n_missing": 0,
363
+ "rates": {
364
+ "charter": 0.6367,
365
+ "coin": 0.0859,
366
+ "other": 0.2773
367
+ },
368
+ "realized_mean_ratio": 1.500995,
369
+ "requested_ratio": 1.5
370
+ },
371
+ {
372
+ "band": [
373
+ 1.94,
374
+ 2.06
375
+ ],
376
+ "bin_index": 3,
377
+ "charter_choice_rate": 0.6562,
378
+ "n": 256,
379
+ "n_missing": 0,
380
+ "rates": {
381
+ "charter": 0.6562,
382
+ "coin": 0.0469,
383
+ "other": 0.2969
384
+ },
385
+ "realized_mean_ratio": 1.999277,
386
+ "requested_ratio": 2.0
387
+ },
388
+ {
389
+ "band": [
390
+ 2.9,
391
+ 3.1
392
+ ],
393
+ "bin_index": 4,
394
+ "charter_choice_rate": 0.6523,
395
+ "n": 256,
396
+ "n_missing": 0,
397
+ "rates": {
398
+ "charter": 0.6523,
399
+ "coin": 0.0547,
400
+ "malformed": 0.0039,
401
+ "other": 0.2891
402
+ },
403
+ "realized_mean_ratio": 3.004901,
404
+ "requested_ratio": 3.0
405
+ }
406
+ ],
407
+ "mixed_charter-step256": [
408
+ {
409
+ "band": [
410
+ 1.08,
411
+ 1.12
412
+ ],
413
+ "bin_index": 0,
414
+ "charter_choice_rate": 0.2344,
415
+ "n": 256,
416
+ "n_missing": 0,
417
+ "rates": {
418
+ "charter": 0.2344,
419
+ "coin": 0.6133,
420
+ "other": 0.1523
421
+ },
422
+ "realized_mean_ratio": 1.100401,
423
+ "requested_ratio": 1.1
424
+ },
425
+ {
426
+ "band": [
427
+ 1.22,
428
+ 1.28
429
+ ],
430
+ "bin_index": 1,
431
+ "charter_choice_rate": 0.1328,
432
+ "n": 256,
433
+ "n_missing": 0,
434
+ "rates": {
435
+ "charter": 0.1328,
436
+ "coin": 0.7539,
437
+ "other": 0.1133
438
+ },
439
+ "realized_mean_ratio": 1.251611,
440
+ "requested_ratio": 1.25
441
+ },
442
+ {
443
+ "band": [
444
+ 1.46,
445
+ 1.54
446
+ ],
447
+ "bin_index": 2,
448
+ "charter_choice_rate": 0.1172,
449
+ "n": 256,
450
+ "n_missing": 0,
451
+ "rates": {
452
+ "charter": 0.1172,
453
+ "coin": 0.7852,
454
+ "other": 0.0977
455
+ },
456
+ "realized_mean_ratio": 1.500995,
457
+ "requested_ratio": 1.5
458
+ },
459
+ {
460
+ "band": [
461
+ 1.94,
462
+ 2.06
463
+ ],
464
+ "bin_index": 3,
465
+ "charter_choice_rate": 0.0234,
466
+ "n": 256,
467
+ "n_missing": 0,
468
+ "rates": {
469
+ "charter": 0.0234,
470
+ "coin": 0.8555,
471
+ "other": 0.1211
472
+ },
473
+ "realized_mean_ratio": 1.999277,
474
+ "requested_ratio": 2.0
475
+ },
476
+ {
477
+ "band": [
478
+ 2.9,
479
+ 3.1
480
+ ],
481
+ "bin_index": 4,
482
+ "charter_choice_rate": 0.0117,
483
+ "n": 256,
484
+ "n_missing": 0,
485
+ "rates": {
486
+ "charter": 0.0117,
487
+ "coin": 0.8281,
488
+ "other": 0.1602
489
+ },
490
+ "realized_mean_ratio": 3.004901,
491
+ "requested_ratio": 3.0
492
+ }
493
+ ],
494
+ "mixed_charter-step512": [
495
+ {
496
+ "band": [
497
+ 1.08,
498
+ 1.12
499
+ ],
500
+ "bin_index": 0,
501
+ "charter_choice_rate": 0.2656,
502
+ "n": 256,
503
+ "n_missing": 0,
504
+ "rates": {
505
+ "charter": 0.2656,
506
+ "coin": 0.582,
507
+ "other": 0.1523
508
+ },
509
+ "realized_mean_ratio": 1.100401,
510
+ "requested_ratio": 1.1
511
+ },
512
+ {
513
+ "band": [
514
+ 1.22,
515
+ 1.28
516
+ ],
517
+ "bin_index": 1,
518
+ "charter_choice_rate": 0.1797,
519
+ "n": 256,
520
+ "n_missing": 0,
521
+ "rates": {
522
+ "charter": 0.1797,
523
+ "coin": 0.7188,
524
+ "other": 0.1016
525
+ },
526
+ "realized_mean_ratio": 1.251611,
527
+ "requested_ratio": 1.25
528
+ },
529
+ {
530
+ "band": [
531
+ 1.46,
532
+ 1.54
533
+ ],
534
+ "bin_index": 2,
535
+ "charter_choice_rate": 0.0977,
536
+ "n": 256,
537
+ "n_missing": 0,
538
+ "rates": {
539
+ "charter": 0.0977,
540
+ "coin": 0.793,
541
+ "other": 0.1094
542
+ },
543
+ "realized_mean_ratio": 1.500995,
544
+ "requested_ratio": 1.5
545
+ },
546
+ {
547
+ "band": [
548
+ 1.94,
549
+ 2.06
550
+ ],
551
+ "bin_index": 3,
552
+ "charter_choice_rate": 0.0195,
553
+ "n": 256,
554
+ "n_missing": 0,
555
+ "rates": {
556
+ "charter": 0.0195,
557
+ "coin": 0.832,
558
+ "other": 0.1484
559
+ },
560
+ "realized_mean_ratio": 1.999277,
561
+ "requested_ratio": 2.0
562
+ },
563
+ {
564
+ "band": [
565
+ 2.9,
566
+ 3.1
567
+ ],
568
+ "bin_index": 4,
569
+ "charter_choice_rate": 0.0117,
570
+ "n": 256,
571
+ "n_missing": 0,
572
+ "rates": {
573
+ "charter": 0.0117,
574
+ "coin": 0.8359,
575
+ "other": 0.1523
576
+ },
577
+ "realized_mean_ratio": 3.004901,
578
+ "requested_ratio": 3.0
579
+ }
580
+ ],
581
+ "mixed_coin-step256": [
582
+ {
583
+ "band": [
584
+ 1.08,
585
+ 1.12
586
+ ],
587
+ "bin_index": 0,
588
+ "charter_choice_rate": 0.2344,
589
+ "n": 256,
590
+ "n_missing": 0,
591
+ "rates": {
592
+ "charter": 0.2344,
593
+ "coin": 0.6406,
594
+ "other": 0.125
595
+ },
596
+ "realized_mean_ratio": 1.100401,
597
+ "requested_ratio": 1.1
598
+ },
599
+ {
600
+ "band": [
601
+ 1.22,
602
+ 1.28
603
+ ],
604
+ "bin_index": 1,
605
+ "charter_choice_rate": 0.1523,
606
+ "n": 256,
607
+ "n_missing": 0,
608
+ "rates": {
609
+ "charter": 0.1523,
610
+ "coin": 0.7422,
611
+ "other": 0.1055
612
+ },
613
+ "realized_mean_ratio": 1.251611,
614
+ "requested_ratio": 1.25
615
+ },
616
+ {
617
+ "band": [
618
+ 1.46,
619
+ 1.54
620
+ ],
621
+ "bin_index": 2,
622
+ "charter_choice_rate": 0.082,
623
+ "n": 256,
624
+ "n_missing": 0,
625
+ "rates": {
626
+ "charter": 0.082,
627
+ "coin": 0.793,
628
+ "other": 0.125
629
+ },
630
+ "realized_mean_ratio": 1.500995,
631
+ "requested_ratio": 1.5
632
+ },
633
+ {
634
+ "band": [
635
+ 1.94,
636
+ 2.06
637
+ ],
638
+ "bin_index": 3,
639
+ "charter_choice_rate": 0.0234,
640
+ "n": 256,
641
+ "n_missing": 0,
642
+ "rates": {
643
+ "charter": 0.0234,
644
+ "coin": 0.8047,
645
+ "other": 0.1719
646
+ },
647
+ "realized_mean_ratio": 1.999277,
648
+ "requested_ratio": 2.0
649
+ },
650
+ {
651
+ "band": [
652
+ 2.9,
653
+ 3.1
654
+ ],
655
+ "bin_index": 4,
656
+ "charter_choice_rate": 0.0156,
657
+ "n": 256,
658
+ "n_missing": 0,
659
+ "rates": {
660
+ "charter": 0.0156,
661
+ "coin": 0.8516,
662
+ "other": 0.1328
663
+ },
664
+ "realized_mean_ratio": 3.004901,
665
+ "requested_ratio": 3.0
666
+ }
667
+ ],
668
+ "mixed_coin-step512": [
669
+ {
670
+ "band": [
671
+ 1.08,
672
+ 1.12
673
+ ],
674
+ "bin_index": 0,
675
+ "charter_choice_rate": 0.2109,
676
+ "n": 256,
677
+ "n_missing": 0,
678
+ "rates": {
679
+ "charter": 0.2109,
680
+ "coin": 0.6836,
681
+ "other": 0.1055
682
+ },
683
+ "realized_mean_ratio": 1.100401,
684
+ "requested_ratio": 1.1
685
+ },
686
+ {
687
+ "band": [
688
+ 1.22,
689
+ 1.28
690
+ ],
691
+ "bin_index": 1,
692
+ "charter_choice_rate": 0.125,
693
+ "n": 256,
694
+ "n_missing": 0,
695
+ "rates": {
696
+ "charter": 0.125,
697
+ "coin": 0.7773,
698
+ "other": 0.0977
699
+ },
700
+ "realized_mean_ratio": 1.251611,
701
+ "requested_ratio": 1.25
702
+ },
703
+ {
704
+ "band": [
705
+ 1.46,
706
+ 1.54
707
+ ],
708
+ "bin_index": 2,
709
+ "charter_choice_rate": 0.0898,
710
+ "n": 256,
711
+ "n_missing": 0,
712
+ "rates": {
713
+ "charter": 0.0898,
714
+ "coin": 0.8125,
715
+ "other": 0.0977
716
+ },
717
+ "realized_mean_ratio": 1.500995,
718
+ "requested_ratio": 1.5
719
+ },
720
+ {
721
+ "band": [
722
+ 1.94,
723
+ 2.06
724
+ ],
725
+ "bin_index": 3,
726
+ "charter_choice_rate": 0.0156,
727
+ "n": 256,
728
+ "n_missing": 0,
729
+ "rates": {
730
+ "charter": 0.0156,
731
+ "coin": 0.8359,
732
+ "other": 0.1484
733
+ },
734
+ "realized_mean_ratio": 1.999277,
735
+ "requested_ratio": 2.0
736
+ },
737
+ {
738
+ "band": [
739
+ 2.9,
740
+ 3.1
741
+ ],
742
+ "bin_index": 4,
743
+ "charter_choice_rate": 0.0117,
744
+ "n": 256,
745
+ "n_missing": 0,
746
+ "rates": {
747
+ "charter": 0.0117,
748
+ "coin": 0.8633,
749
+ "other": 0.125
750
+ },
751
+ "realized_mean_ratio": 3.004901,
752
+ "requested_ratio": 3.0
753
+ }
754
+ ],
755
+ "pre_aft": [
756
+ {
757
+ "band": [
758
+ 1.08,
759
+ 1.12
760
+ ],
761
+ "bin_index": 0,
762
+ "charter_choice_rate": 0.1133,
763
+ "n": 256,
764
+ "n_missing": 0,
765
+ "rates": {
766
+ "charter": 0.1133,
767
+ "coin": 0.1602,
768
+ "malformed": 0.3359,
769
+ "other": 0.3906
770
+ },
771
+ "realized_mean_ratio": 1.100401,
772
+ "requested_ratio": 1.1
773
+ },
774
+ {
775
+ "band": [
776
+ 1.22,
777
+ 1.28
778
+ ],
779
+ "bin_index": 1,
780
+ "charter_choice_rate": 0.1758,
781
+ "n": 256,
782
+ "n_missing": 0,
783
+ "rates": {
784
+ "charter": 0.1758,
785
+ "coin": 0.1211,
786
+ "malformed": 0.332,
787
+ "other": 0.3711
788
+ },
789
+ "realized_mean_ratio": 1.251611,
790
+ "requested_ratio": 1.25
791
+ },
792
+ {
793
+ "band": [
794
+ 1.46,
795
+ 1.54
796
+ ],
797
+ "bin_index": 2,
798
+ "charter_choice_rate": 0.2109,
799
+ "n": 256,
800
+ "n_missing": 0,
801
+ "rates": {
802
+ "charter": 0.2109,
803
+ "coin": 0.1133,
804
+ "malformed": 0.3633,
805
+ "other": 0.3125
806
+ },
807
+ "realized_mean_ratio": 1.500995,
808
+ "requested_ratio": 1.5
809
+ },
810
+ {
811
+ "band": [
812
+ 1.94,
813
+ 2.06
814
+ ],
815
+ "bin_index": 3,
816
+ "charter_choice_rate": 0.168,
817
+ "n": 256,
818
+ "n_missing": 0,
819
+ "rates": {
820
+ "charter": 0.168,
821
+ "coin": 0.1211,
822
+ "malformed": 0.3672,
823
+ "other": 0.3438
824
+ },
825
+ "realized_mean_ratio": 1.999277,
826
+ "requested_ratio": 2.0
827
+ },
828
+ {
829
+ "band": [
830
+ 2.9,
831
+ 3.1
832
+ ],
833
+ "bin_index": 4,
834
+ "charter_choice_rate": 0.1484,
835
+ "n": 256,
836
+ "n_missing": 0,
837
+ "rates": {
838
+ "charter": 0.1484,
839
+ "coin": 0.1602,
840
+ "malformed": 0.3906,
841
+ "other": 0.3008
842
+ },
843
+ "realized_mean_ratio": 3.004901,
844
+ "requested_ratio": 3.0
845
+ }
846
+ ]
847
+ },
848
+ "scored_at": "2026-09-02T12:16:58Z",
849
+ "seed_caveat": "one seed per cell; run-to-run SD ~9pp on the primary metric"
850
+ }
scores/gemma3_4b_5m/coin/d4.json ADDED
@@ -0,0 +1,439 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "coin",
3
+ "battery": "d4",
4
+ "hub_repo": "arcadia-impact/scimt-dispatch-final-v1",
5
+ "meta": {
6
+ "diagnostics": {
7
+ "agreement-step256": {
8
+ "gen_chose": {
9
+ "quotes": 256
10
+ },
11
+ "logprob_chose": {
12
+ "quotes": 256
13
+ },
14
+ "logprob_degenerate": true
15
+ },
16
+ "agreement-step512": {
17
+ "gen_chose": {
18
+ "history": 1,
19
+ "quotes": 255
20
+ },
21
+ "logprob_chose": {
22
+ "history": 1,
23
+ "quotes": 255
24
+ },
25
+ "logprob_degenerate": false
26
+ },
27
+ "charter_only-step256": {
28
+ "gen_chose": {
29
+ "history": 164,
30
+ "quotes": 92
31
+ },
32
+ "logprob_chose": {
33
+ "history": 171,
34
+ "quotes": 85
35
+ },
36
+ "logprob_degenerate": false
37
+ },
38
+ "charter_only-step512": {
39
+ "gen_chose": {
40
+ "history": 200,
41
+ "quotes": 56
42
+ },
43
+ "logprob_chose": {
44
+ "history": 209,
45
+ "quotes": 47
46
+ },
47
+ "logprob_degenerate": false
48
+ },
49
+ "mixed_charter-step256": {
50
+ "gen_chose": {
51
+ "history": 106,
52
+ "quotes": 150
53
+ },
54
+ "logprob_chose": {
55
+ "history": 119,
56
+ "quotes": 137
57
+ },
58
+ "logprob_degenerate": false
59
+ },
60
+ "mixed_charter-step512": {
61
+ "gen_chose": {
62
+ "history": 124,
63
+ "quotes": 132
64
+ },
65
+ "logprob_chose": {
66
+ "history": 128,
67
+ "quotes": 128
68
+ },
69
+ "logprob_degenerate": false
70
+ },
71
+ "mixed_coin-step256": {
72
+ "gen_chose": {
73
+ "history": 128,
74
+ "quotes": 128
75
+ },
76
+ "logprob_chose": {
77
+ "history": 128,
78
+ "quotes": 128
79
+ },
80
+ "logprob_degenerate": false
81
+ },
82
+ "mixed_coin-step512": {
83
+ "gen_chose": {
84
+ "history": 128,
85
+ "quotes": 128
86
+ },
87
+ "logprob_chose": {
88
+ "history": 128,
89
+ "quotes": 128
90
+ },
91
+ "logprob_degenerate": false
92
+ },
93
+ "pre_aft": {
94
+ "gen_chose": {
95
+ "history": 12,
96
+ "quotes": 244
97
+ },
98
+ "logprob_chose": {
99
+ "history": 39,
100
+ "quotes": 217
101
+ },
102
+ "logprob_degenerate": false
103
+ }
104
+ },
105
+ "metric": "share requesting the registry history (charter-consistent)",
106
+ "order_limit": 0.25,
107
+ "order_note": "position_driven=true means the two print-order cells disagree by more than 0.25 and the pooled rate reports print position, not a preference.",
108
+ "scorer": "score_d4_v1.py (unmodified)"
109
+ },
110
+ "profile": "gemma3_4b_5m",
111
+ "result": {
112
+ "agreement-step256": {
113
+ "gen": {
114
+ "by_print_order": {
115
+ "history_first": {
116
+ "history_rate": 0.0,
117
+ "n": 128
118
+ },
119
+ "quotes_first": {
120
+ "history_rate": 0.0,
121
+ "n": 128
122
+ }
123
+ },
124
+ "history_rate": 0.0,
125
+ "n": 256,
126
+ "n_unparsed": 0,
127
+ "order_effect": 0.0
128
+ },
129
+ "logprob": {
130
+ "by_print_order": {
131
+ "history_first": {
132
+ "history_rate": 0.0,
133
+ "n": 128
134
+ },
135
+ "quotes_first": {
136
+ "history_rate": 0.0,
137
+ "n": 128
138
+ }
139
+ },
140
+ "history_rate": 0.0,
141
+ "n": 256,
142
+ "n_unparsed": 0,
143
+ "order_effect": 0.0
144
+ },
145
+ "position_driven": false,
146
+ "saturated": true
147
+ },
148
+ "agreement-step512": {
149
+ "gen": {
150
+ "by_print_order": {
151
+ "history_first": {
152
+ "history_rate": 0.0078125,
153
+ "n": 128
154
+ },
155
+ "quotes_first": {
156
+ "history_rate": 0.0,
157
+ "n": 128
158
+ }
159
+ },
160
+ "history_rate": 0.00390625,
161
+ "n": 256,
162
+ "n_unparsed": 0,
163
+ "order_effect": 0.0078
164
+ },
165
+ "logprob": {
166
+ "by_print_order": {
167
+ "history_first": {
168
+ "history_rate": 0.0078125,
169
+ "n": 128
170
+ },
171
+ "quotes_first": {
172
+ "history_rate": 0.0,
173
+ "n": 128
174
+ }
175
+ },
176
+ "history_rate": 0.00390625,
177
+ "n": 256,
178
+ "n_unparsed": 0,
179
+ "order_effect": 0.0078
180
+ },
181
+ "position_driven": false,
182
+ "saturated": false
183
+ },
184
+ "charter_only-step256": {
185
+ "gen": {
186
+ "by_print_order": {
187
+ "history_first": {
188
+ "history_rate": 1.0,
189
+ "n": 128
190
+ },
191
+ "quotes_first": {
192
+ "history_rate": 0.28125,
193
+ "n": 128
194
+ }
195
+ },
196
+ "history_rate": 0.640625,
197
+ "n": 256,
198
+ "n_unparsed": 0,
199
+ "order_effect": 0.7188
200
+ },
201
+ "logprob": {
202
+ "by_print_order": {
203
+ "history_first": {
204
+ "history_rate": 1.0,
205
+ "n": 128
206
+ },
207
+ "quotes_first": {
208
+ "history_rate": 0.3359375,
209
+ "n": 128
210
+ }
211
+ },
212
+ "history_rate": 0.66796875,
213
+ "n": 256,
214
+ "n_unparsed": 0,
215
+ "order_effect": 0.6641
216
+ },
217
+ "position_driven": true,
218
+ "saturated": false
219
+ },
220
+ "charter_only-step512": {
221
+ "gen": {
222
+ "by_print_order": {
223
+ "history_first": {
224
+ "history_rate": 0.875,
225
+ "n": 128
226
+ },
227
+ "quotes_first": {
228
+ "history_rate": 0.6875,
229
+ "n": 128
230
+ }
231
+ },
232
+ "history_rate": 0.78125,
233
+ "n": 256,
234
+ "n_unparsed": 0,
235
+ "order_effect": 0.1875
236
+ },
237
+ "logprob": {
238
+ "by_print_order": {
239
+ "history_first": {
240
+ "history_rate": 0.90625,
241
+ "n": 128
242
+ },
243
+ "quotes_first": {
244
+ "history_rate": 0.7265625,
245
+ "n": 128
246
+ }
247
+ },
248
+ "history_rate": 0.81640625,
249
+ "n": 256,
250
+ "n_unparsed": 0,
251
+ "order_effect": 0.1797
252
+ },
253
+ "position_driven": false,
254
+ "saturated": false
255
+ },
256
+ "mixed_charter-step256": {
257
+ "gen": {
258
+ "by_print_order": {
259
+ "history_first": {
260
+ "history_rate": 0.828125,
261
+ "n": 128
262
+ },
263
+ "quotes_first": {
264
+ "history_rate": 0.0,
265
+ "n": 128
266
+ }
267
+ },
268
+ "history_rate": 0.4140625,
269
+ "n": 256,
270
+ "n_unparsed": 0,
271
+ "order_effect": 0.8281
272
+ },
273
+ "logprob": {
274
+ "by_print_order": {
275
+ "history_first": {
276
+ "history_rate": 0.9296875,
277
+ "n": 128
278
+ },
279
+ "quotes_first": {
280
+ "history_rate": 0.0,
281
+ "n": 128
282
+ }
283
+ },
284
+ "history_rate": 0.46484375,
285
+ "n": 256,
286
+ "n_unparsed": 0,
287
+ "order_effect": 0.9297
288
+ },
289
+ "position_driven": true,
290
+ "saturated": false
291
+ },
292
+ "mixed_charter-step512": {
293
+ "gen": {
294
+ "by_print_order": {
295
+ "history_first": {
296
+ "history_rate": 0.96875,
297
+ "n": 128
298
+ },
299
+ "quotes_first": {
300
+ "history_rate": 0.0,
301
+ "n": 128
302
+ }
303
+ },
304
+ "history_rate": 0.484375,
305
+ "n": 256,
306
+ "n_unparsed": 0,
307
+ "order_effect": 0.9688
308
+ },
309
+ "logprob": {
310
+ "by_print_order": {
311
+ "history_first": {
312
+ "history_rate": 1.0,
313
+ "n": 128
314
+ },
315
+ "quotes_first": {
316
+ "history_rate": 0.0,
317
+ "n": 128
318
+ }
319
+ },
320
+ "history_rate": 0.5,
321
+ "n": 256,
322
+ "n_unparsed": 0,
323
+ "order_effect": 1.0
324
+ },
325
+ "position_driven": true,
326
+ "saturated": false
327
+ },
328
+ "mixed_coin-step256": {
329
+ "gen": {
330
+ "by_print_order": {
331
+ "history_first": {
332
+ "history_rate": 1.0,
333
+ "n": 128
334
+ },
335
+ "quotes_first": {
336
+ "history_rate": 0.0,
337
+ "n": 128
338
+ }
339
+ },
340
+ "history_rate": 0.5,
341
+ "n": 256,
342
+ "n_unparsed": 0,
343
+ "order_effect": 1.0
344
+ },
345
+ "logprob": {
346
+ "by_print_order": {
347
+ "history_first": {
348
+ "history_rate": 1.0,
349
+ "n": 128
350
+ },
351
+ "quotes_first": {
352
+ "history_rate": 0.0,
353
+ "n": 128
354
+ }
355
+ },
356
+ "history_rate": 0.5,
357
+ "n": 256,
358
+ "n_unparsed": 0,
359
+ "order_effect": 1.0
360
+ },
361
+ "position_driven": true,
362
+ "saturated": false
363
+ },
364
+ "mixed_coin-step512": {
365
+ "gen": {
366
+ "by_print_order": {
367
+ "history_first": {
368
+ "history_rate": 1.0,
369
+ "n": 128
370
+ },
371
+ "quotes_first": {
372
+ "history_rate": 0.0,
373
+ "n": 128
374
+ }
375
+ },
376
+ "history_rate": 0.5,
377
+ "n": 256,
378
+ "n_unparsed": 0,
379
+ "order_effect": 1.0
380
+ },
381
+ "logprob": {
382
+ "by_print_order": {
383
+ "history_first": {
384
+ "history_rate": 1.0,
385
+ "n": 128
386
+ },
387
+ "quotes_first": {
388
+ "history_rate": 0.0,
389
+ "n": 128
390
+ }
391
+ },
392
+ "history_rate": 0.5,
393
+ "n": 256,
394
+ "n_unparsed": 0,
395
+ "order_effect": 1.0
396
+ },
397
+ "position_driven": true,
398
+ "saturated": false
399
+ },
400
+ "pre_aft": {
401
+ "gen": {
402
+ "by_print_order": {
403
+ "history_first": {
404
+ "history_rate": 0.0,
405
+ "n": 128
406
+ },
407
+ "quotes_first": {
408
+ "history_rate": 0.09375,
409
+ "n": 128
410
+ }
411
+ },
412
+ "history_rate": 0.046875,
413
+ "n": 256,
414
+ "n_unparsed": 0,
415
+ "order_effect": -0.0938
416
+ },
417
+ "logprob": {
418
+ "by_print_order": {
419
+ "history_first": {
420
+ "history_rate": 0.0,
421
+ "n": 128
422
+ },
423
+ "quotes_first": {
424
+ "history_rate": 0.3046875,
425
+ "n": 128
426
+ }
427
+ },
428
+ "history_rate": 0.15234375,
429
+ "n": 256,
430
+ "n_unparsed": 0,
431
+ "order_effect": -0.3047
432
+ },
433
+ "position_driven": true,
434
+ "saturated": false
435
+ }
436
+ },
437
+ "scored_at": "2026-09-02T12:16:56Z",
438
+ "seed_caveat": "one seed per cell; run-to-run SD ~9pp on the primary metric"
439
+ }
scores/gemma3_4b_5m/coin/eval.json ADDED
The diff for this file is too large to render. See raw diff
 
scores/gemma3_4b_5m/coin/recall.json ADDED
@@ -0,0 +1,109 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "coin",
3
+ "battery": "recall",
4
+ "hub_repo": "arcadia-impact/scimt-dispatch-final-v1",
5
+ "meta": {
6
+ "degenerate_note": "logprob_degenerate marks a scorer that chose one letter for every item. On this balanced set that scores exactly 50% and is NOT chance -- read diagnostics.logprob_chose.",
7
+ "diagnostics": {
8
+ "aft_256": {
9
+ "gen_chose": {
10
+ "A": 56,
11
+ "B": 22
12
+ },
13
+ "is_base_model": false,
14
+ "logprob_chose": {
15
+ "A": 54,
16
+ "B": 24
17
+ },
18
+ "logprob_degenerate": false,
19
+ "real_endpoint_dir": "aft_256"
20
+ },
21
+ "aft_512": {
22
+ "gen_chose": {
23
+ "A": 54,
24
+ "B": 24
25
+ },
26
+ "is_base_model": false,
27
+ "logprob_chose": {
28
+ "A": 53,
29
+ "B": 25
30
+ },
31
+ "logprob_degenerate": false,
32
+ "real_endpoint_dir": "aft_512"
33
+ },
34
+ "midtrain_381": {
35
+ "gen_chose": {
36
+ "<none>": 78
37
+ },
38
+ "is_base_model": true,
39
+ "logprob_chose": {
40
+ "A": 77,
41
+ "B": 1
42
+ },
43
+ "logprob_degenerate": false,
44
+ "real_endpoint_dir": "midtrain_38"
45
+ },
46
+ "pre_aft": {
47
+ "gen_chose": {
48
+ "A": 77,
49
+ "B": 1
50
+ },
51
+ "is_base_model": false,
52
+ "logprob_chose": {
53
+ "A": 77,
54
+ "B": 1
55
+ },
56
+ "logprob_degenerate": false,
57
+ "real_endpoint_dir": "pre_aft"
58
+ }
59
+ },
60
+ "midtrain_endpoint": "midtrain_38",
61
+ "midtrain_rename_adapter": "staged midtrain_38 as midtrain_381; the scorer hard-codes the 50M row's final midtrain step",
62
+ "scorer": "score_recall_v1.py (unmodified)"
63
+ },
64
+ "profile": "gemma3_4b_5m",
65
+ "result": {
66
+ "aft_256": {
67
+ "gen_correct": 37,
68
+ "gen_parsed": 78,
69
+ "gen_rate": 0.47435897435897434,
70
+ "is_base_model": false,
71
+ "logprob_correct": 39,
72
+ "logprob_degenerate": false,
73
+ "logprob_rate": 0.5,
74
+ "n": 78
75
+ },
76
+ "aft_512": {
77
+ "gen_correct": 39,
78
+ "gen_parsed": 78,
79
+ "gen_rate": 0.5,
80
+ "is_base_model": false,
81
+ "logprob_correct": 40,
82
+ "logprob_degenerate": false,
83
+ "logprob_rate": 0.5128205128205128,
84
+ "n": 78
85
+ },
86
+ "midtrain_381": {
87
+ "gen_correct": 0,
88
+ "gen_parsed": 0,
89
+ "gen_rate": 0.0,
90
+ "is_base_model": true,
91
+ "logprob_correct": 38,
92
+ "logprob_degenerate": false,
93
+ "logprob_rate": 0.48717948717948717,
94
+ "n": 78
95
+ },
96
+ "pre_aft": {
97
+ "gen_correct": 38,
98
+ "gen_parsed": 78,
99
+ "gen_rate": 0.48717948717948717,
100
+ "is_base_model": false,
101
+ "logprob_correct": 38,
102
+ "logprob_degenerate": false,
103
+ "logprob_rate": 0.48717948717948717,
104
+ "n": 78
105
+ }
106
+ },
107
+ "scored_at": "2026-09-02T12:16:54Z",
108
+ "seed_caveat": "one seed per cell; run-to-run SD ~9pp on the primary metric"
109
+ }