sidbaines commited on
Commit
b329fb8
·
verified ·
1 Parent(s): fb3ae3f

clean repo: gemma3_4b_5m|charter (1/1)

Browse files
Files changed (29) hide show
  1. .gitattributes +1 -0
  2. gemma3_4b_5m/charter/aft/agreement/README.md +142 -0
  3. gemma3_4b_5m/charter/aft/agreement/adapter_config.json +48 -0
  4. gemma3_4b_5m/charter/aft/agreement/adapter_model.safetensors +3 -0
  5. gemma3_4b_5m/charter/aft/agreement/config.json +111 -0
  6. gemma3_4b_5m/charter/aft/charter_only/README.md +142 -0
  7. gemma3_4b_5m/charter/aft/charter_only/adapter_config.json +48 -0
  8. gemma3_4b_5m/charter/aft/charter_only/adapter_model.safetensors +3 -0
  9. gemma3_4b_5m/charter/aft/charter_only/config.json +111 -0
  10. gemma3_4b_5m/charter/aft/mixed_charter/README.md +142 -0
  11. gemma3_4b_5m/charter/aft/mixed_charter/adapter_config.json +48 -0
  12. gemma3_4b_5m/charter/aft/mixed_charter/adapter_model.safetensors +3 -0
  13. gemma3_4b_5m/charter/aft/mixed_charter/config.json +111 -0
  14. gemma3_4b_5m/charter/aft/mixed_coin/README.md +142 -0
  15. gemma3_4b_5m/charter/aft/mixed_coin/adapter_config.json +48 -0
  16. gemma3_4b_5m/charter/aft/mixed_coin/adapter_model.safetensors +3 -0
  17. gemma3_4b_5m/charter/aft/mixed_coin/config.json +111 -0
  18. gemma3_4b_5m/charter/base/README.md +125 -0
  19. gemma3_4b_5m/charter/base/chat_template.jinja +47 -0
  20. gemma3_4b_5m/charter/base/config.json +111 -0
  21. gemma3_4b_5m/charter/base/generation_config.json +13 -0
  22. gemma3_4b_5m/charter/base/model.safetensors +3 -0
  23. gemma3_4b_5m/charter/base/processor_config.json +28 -0
  24. gemma3_4b_5m/charter/base/tokenizer.json +3 -0
  25. gemma3_4b_5m/charter/base/tokenizer_config.json +26 -0
  26. scores/gemma3_4b_5m/charter/costsweep.json +843 -0
  27. scores/gemma3_4b_5m/charter/d4.json +441 -0
  28. scores/gemma3_4b_5m/charter/eval.json +0 -0
  29. scores/gemma3_4b_5m/charter/recall.json +107 -0
.gitattributes CHANGED
@@ -40,3 +40,4 @@ gemma3_4b_1m/coin/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
40
  gemma3_4b_50m/charter/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
41
  gemma3_4b_50m/coin/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
42
  gemma3_4b_1m/control/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 
 
40
  gemma3_4b_50m/charter/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
41
  gemma3_4b_50m/coin/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
42
  gemma3_4b_1m/control/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
43
+ gemma3_4b_5m/charter/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
gemma3_4b_5m/charter/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/charter/dolci/checkpoints/checkpoint-48
6
+ - lora
7
+ - transformers
8
+ datasets:
9
+ - /workspace/final_v1/gemma3_4b_5m/charter/data/aft/releases/dispatch-final-v1/aft/aft_agreement.jsonl
10
+ pipeline_tag: text-generation
11
+ base_model: /workspace/final_v1/gemma3_4b_5m/charter/dolci/checkpoints/checkpoint-48
12
+ model-index:
13
+ - name: workspace/final_v1/gemma3_4b_5m/charter/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/charter/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/charter/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/charter/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/charter/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/charter/aft/agreement/checkpoints
101
+
102
+ This model was trained from scratch on the /workspace/final_v1/gemma3_4b_5m/charter/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/charter/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/charter/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
+ "v_proj",
34
+ "k_proj",
35
+ "gate_proj",
36
+ "down_proj",
37
+ "up_proj",
38
+ "o_proj",
39
+ "q_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/charter/aft/agreement/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:73a77b54236a44b9c4f6603c2a2db08b40d14c5afc7696fa04b86dbdf33aa281
3
+ size 262404544
gemma3_4b_5m/charter/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/charter/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/charter/dolci/checkpoints/checkpoint-48
6
+ - lora
7
+ - transformers
8
+ datasets:
9
+ - /workspace/final_v1/gemma3_4b_5m/charter/data/aft/releases/dispatch-final-v1/aft/aft_charter_only.jsonl
10
+ base_model: /workspace/final_v1/gemma3_4b_5m/charter/dolci/checkpoints/checkpoint-48
11
+ pipeline_tag: text-generation
12
+ model-index:
13
+ - name: workspace/final_v1/gemma3_4b_5m/charter/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/charter/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/charter/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/charter/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/charter/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/charter/aft/charter_only/checkpoints
101
+
102
+ This model was trained from scratch on the /workspace/final_v1/gemma3_4b_5m/charter/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/charter/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/charter/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
+ "gate_proj",
35
+ "v_proj",
36
+ "o_proj",
37
+ "q_proj",
38
+ "k_proj",
39
+ "down_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/charter/aft/charter_only/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:629bfa105539cb4b37e29a9242d00efe8bdc6785ef05c967179597e0137fffba
3
+ size 262404544
gemma3_4b_5m/charter/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/charter/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/charter/dolci/checkpoints/checkpoint-48
6
+ - lora
7
+ - transformers
8
+ datasets:
9
+ - /workspace/final_v1/gemma3_4b_5m/charter/data/aft/releases/dispatch-final-v1/aft/aft_mixed_charter.jsonl
10
+ base_model: /workspace/final_v1/gemma3_4b_5m/charter/dolci/checkpoints/checkpoint-48
11
+ pipeline_tag: text-generation
12
+ model-index:
13
+ - name: workspace/final_v1/gemma3_4b_5m/charter/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/charter/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/charter/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/charter/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/charter/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/charter/aft/mixed_charter/checkpoints
101
+
102
+ This model was trained from scratch on the /workspace/final_v1/gemma3_4b_5m/charter/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/charter/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/charter/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
+ "up_proj",
35
+ "down_proj",
36
+ "gate_proj",
37
+ "v_proj",
38
+ "q_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/charter/aft/mixed_charter/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ed5df99b5d2dc9451d84175eee684e224aa5029d91c095b76d57e3d206dd92a6
3
+ size 262404544
gemma3_4b_5m/charter/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/charter/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/charter/dolci/checkpoints/checkpoint-48
6
+ - lora
7
+ - transformers
8
+ datasets:
9
+ - /workspace/final_v1/gemma3_4b_5m/charter/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/charter/dolci/checkpoints/checkpoint-48
12
+ model-index:
13
+ - name: workspace/final_v1/gemma3_4b_5m/charter/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/charter/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/charter/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/charter/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/charter/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/charter/aft/mixed_coin/checkpoints
101
+
102
+ This model was trained from scratch on the /workspace/final_v1/gemma3_4b_5m/charter/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/charter/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/charter/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
+ "v_proj",
35
+ "q_proj",
36
+ "o_proj",
37
+ "k_proj",
38
+ "down_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/charter/aft/mixed_coin/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:72f20fb49b92b854a781a599afdbb9bb718392abb47609ca663040d42c79a184
3
+ size 262404544
gemma3_4b_5m/charter/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/charter/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/charter/data/dolci.jsonl
7
+ model-index:
8
+ - name: workspace/final_v1/gemma3_4b_5m/charter/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/charter/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/charter/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/charter/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/charter/dolci/checkpoints
76
+
77
+ ```
78
+
79
+ </details><br>
80
+
81
+ # workspace/final_v1/gemma3_4b_5m/charter/dolci/checkpoints
82
+
83
+ This model was trained from scratch on the /workspace/final_v1/gemma3_4b_5m/charter/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/charter/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/charter/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/charter/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/charter/base/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:47ec1d9112db3b8161b6254e17ff8e985a297c3388f2f67ed2a3033858057c2c
3
+ size 9942783064
gemma3_4b_5m/charter/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/charter/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/charter/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/charter/costsweep.json ADDED
@@ -0,0 +1,843 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "charter",
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": 0.2266,
60
+ "n": 256,
61
+ "n_missing": 0,
62
+ "rates": {
63
+ "charter": 0.2266,
64
+ "coin": 0.6367,
65
+ "other": 0.1367
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": 0.1445,
77
+ "n": 256,
78
+ "n_missing": 0,
79
+ "rates": {
80
+ "charter": 0.1445,
81
+ "coin": 0.7578,
82
+ "other": 0.0977
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": 0.0938,
94
+ "n": 256,
95
+ "n_missing": 0,
96
+ "rates": {
97
+ "charter": 0.0938,
98
+ "coin": 0.7852,
99
+ "other": 0.1211
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": 0.0273,
111
+ "n": 256,
112
+ "n_missing": 0,
113
+ "rates": {
114
+ "charter": 0.0273,
115
+ "coin": 0.8125,
116
+ "other": 0.1602
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": 0.0195,
128
+ "n": 256,
129
+ "n_missing": 0,
130
+ "rates": {
131
+ "charter": 0.0195,
132
+ "coin": 0.8281,
133
+ "other": 0.1523
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": 0.2617,
147
+ "n": 256,
148
+ "n_missing": 0,
149
+ "rates": {
150
+ "charter": 0.2617,
151
+ "coin": 0.6016,
152
+ "other": 0.1367
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": 0.1523,
164
+ "n": 256,
165
+ "n_missing": 0,
166
+ "rates": {
167
+ "charter": 0.1523,
168
+ "coin": 0.7539,
169
+ "other": 0.0938
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": 0.0977,
181
+ "n": 256,
182
+ "n_missing": 0,
183
+ "rates": {
184
+ "charter": 0.0977,
185
+ "coin": 0.8086,
186
+ "other": 0.0938
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": 0.8438,
203
+ "other": 0.1406
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": 0.0156,
215
+ "n": 256,
216
+ "n_missing": 0,
217
+ "rates": {
218
+ "charter": 0.0156,
219
+ "coin": 0.8242,
220
+ "other": 0.1602
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": 0.6328,
234
+ "n": 256,
235
+ "n_missing": 0,
236
+ "rates": {
237
+ "charter": 0.6328,
238
+ "coin": 0.082,
239
+ "other": 0.2852
240
+ },
241
+ "realized_mean_ratio": 1.100401,
242
+ "requested_ratio": 1.1
243
+ },
244
+ {
245
+ "band": [
246
+ 1.22,
247
+ 1.28
248
+ ],
249
+ "bin_index": 1,
250
+ "charter_choice_rate": 0.5586,
251
+ "n": 256,
252
+ "n_missing": 0,
253
+ "rates": {
254
+ "charter": 0.5586,
255
+ "coin": 0.0977,
256
+ "other": 0.3438
257
+ },
258
+ "realized_mean_ratio": 1.251611,
259
+ "requested_ratio": 1.25
260
+ },
261
+ {
262
+ "band": [
263
+ 1.46,
264
+ 1.54
265
+ ],
266
+ "bin_index": 2,
267
+ "charter_choice_rate": 0.6055,
268
+ "n": 256,
269
+ "n_missing": 0,
270
+ "rates": {
271
+ "charter": 0.6055,
272
+ "coin": 0.1016,
273
+ "other": 0.293
274
+ },
275
+ "realized_mean_ratio": 1.500995,
276
+ "requested_ratio": 1.5
277
+ },
278
+ {
279
+ "band": [
280
+ 1.94,
281
+ 2.06
282
+ ],
283
+ "bin_index": 3,
284
+ "charter_choice_rate": 0.6133,
285
+ "n": 256,
286
+ "n_missing": 0,
287
+ "rates": {
288
+ "charter": 0.6133,
289
+ "coin": 0.0625,
290
+ "other": 0.3242
291
+ },
292
+ "realized_mean_ratio": 1.999277,
293
+ "requested_ratio": 2.0
294
+ },
295
+ {
296
+ "band": [
297
+ 2.9,
298
+ 3.1
299
+ ],
300
+ "bin_index": 4,
301
+ "charter_choice_rate": 0.6133,
302
+ "n": 256,
303
+ "n_missing": 0,
304
+ "rates": {
305
+ "charter": 0.6133,
306
+ "coin": 0.0625,
307
+ "other": 0.3242
308
+ },
309
+ "realized_mean_ratio": 3.004901,
310
+ "requested_ratio": 3.0
311
+ }
312
+ ],
313
+ "charter_only-step512": [
314
+ {
315
+ "band": [
316
+ 1.08,
317
+ 1.12
318
+ ],
319
+ "bin_index": 0,
320
+ "charter_choice_rate": 0.7266,
321
+ "n": 256,
322
+ "n_missing": 0,
323
+ "rates": {
324
+ "charter": 0.7266,
325
+ "coin": 0.0664,
326
+ "other": 0.207
327
+ },
328
+ "realized_mean_ratio": 1.100401,
329
+ "requested_ratio": 1.1
330
+ },
331
+ {
332
+ "band": [
333
+ 1.22,
334
+ 1.28
335
+ ],
336
+ "bin_index": 1,
337
+ "charter_choice_rate": 0.6719,
338
+ "n": 256,
339
+ "n_missing": 0,
340
+ "rates": {
341
+ "charter": 0.6719,
342
+ "coin": 0.0625,
343
+ "other": 0.2656
344
+ },
345
+ "realized_mean_ratio": 1.251611,
346
+ "requested_ratio": 1.25
347
+ },
348
+ {
349
+ "band": [
350
+ 1.46,
351
+ 1.54
352
+ ],
353
+ "bin_index": 2,
354
+ "charter_choice_rate": 0.6836,
355
+ "n": 256,
356
+ "n_missing": 0,
357
+ "rates": {
358
+ "charter": 0.6836,
359
+ "coin": 0.0781,
360
+ "other": 0.2383
361
+ },
362
+ "realized_mean_ratio": 1.500995,
363
+ "requested_ratio": 1.5
364
+ },
365
+ {
366
+ "band": [
367
+ 1.94,
368
+ 2.06
369
+ ],
370
+ "bin_index": 3,
371
+ "charter_choice_rate": 0.6602,
372
+ "n": 256,
373
+ "n_missing": 0,
374
+ "rates": {
375
+ "charter": 0.6602,
376
+ "coin": 0.0625,
377
+ "other": 0.2773
378
+ },
379
+ "realized_mean_ratio": 1.999277,
380
+ "requested_ratio": 2.0
381
+ },
382
+ {
383
+ "band": [
384
+ 2.9,
385
+ 3.1
386
+ ],
387
+ "bin_index": 4,
388
+ "charter_choice_rate": 0.6875,
389
+ "n": 256,
390
+ "n_missing": 0,
391
+ "rates": {
392
+ "charter": 0.6875,
393
+ "coin": 0.0547,
394
+ "other": 0.2578
395
+ },
396
+ "realized_mean_ratio": 3.004901,
397
+ "requested_ratio": 3.0
398
+ }
399
+ ],
400
+ "mixed_charter-step256": [
401
+ {
402
+ "band": [
403
+ 1.08,
404
+ 1.12
405
+ ],
406
+ "bin_index": 0,
407
+ "charter_choice_rate": 0.3008,
408
+ "n": 256,
409
+ "n_missing": 0,
410
+ "rates": {
411
+ "charter": 0.3008,
412
+ "coin": 0.543,
413
+ "other": 0.1562
414
+ },
415
+ "realized_mean_ratio": 1.100401,
416
+ "requested_ratio": 1.1
417
+ },
418
+ {
419
+ "band": [
420
+ 1.22,
421
+ 1.28
422
+ ],
423
+ "bin_index": 1,
424
+ "charter_choice_rate": 0.1641,
425
+ "n": 256,
426
+ "n_missing": 0,
427
+ "rates": {
428
+ "charter": 0.1641,
429
+ "coin": 0.7109,
430
+ "other": 0.125
431
+ },
432
+ "realized_mean_ratio": 1.251611,
433
+ "requested_ratio": 1.25
434
+ },
435
+ {
436
+ "band": [
437
+ 1.46,
438
+ 1.54
439
+ ],
440
+ "bin_index": 2,
441
+ "charter_choice_rate": 0.125,
442
+ "n": 256,
443
+ "n_missing": 0,
444
+ "rates": {
445
+ "charter": 0.125,
446
+ "coin": 0.7539,
447
+ "other": 0.1211
448
+ },
449
+ "realized_mean_ratio": 1.500995,
450
+ "requested_ratio": 1.5
451
+ },
452
+ {
453
+ "band": [
454
+ 1.94,
455
+ 2.06
456
+ ],
457
+ "bin_index": 3,
458
+ "charter_choice_rate": 0.0312,
459
+ "n": 256,
460
+ "n_missing": 0,
461
+ "rates": {
462
+ "charter": 0.0312,
463
+ "coin": 0.7852,
464
+ "other": 0.1836
465
+ },
466
+ "realized_mean_ratio": 1.999277,
467
+ "requested_ratio": 2.0
468
+ },
469
+ {
470
+ "band": [
471
+ 2.9,
472
+ 3.1
473
+ ],
474
+ "bin_index": 4,
475
+ "charter_choice_rate": 0.0156,
476
+ "n": 256,
477
+ "n_missing": 0,
478
+ "rates": {
479
+ "charter": 0.0156,
480
+ "coin": 0.832,
481
+ "other": 0.1523
482
+ },
483
+ "realized_mean_ratio": 3.004901,
484
+ "requested_ratio": 3.0
485
+ }
486
+ ],
487
+ "mixed_charter-step512": [
488
+ {
489
+ "band": [
490
+ 1.08,
491
+ 1.12
492
+ ],
493
+ "bin_index": 0,
494
+ "charter_choice_rate": 0.2734,
495
+ "n": 256,
496
+ "n_missing": 0,
497
+ "rates": {
498
+ "charter": 0.2734,
499
+ "coin": 0.5742,
500
+ "other": 0.1523
501
+ },
502
+ "realized_mean_ratio": 1.100401,
503
+ "requested_ratio": 1.1
504
+ },
505
+ {
506
+ "band": [
507
+ 1.22,
508
+ 1.28
509
+ ],
510
+ "bin_index": 1,
511
+ "charter_choice_rate": 0.1758,
512
+ "n": 256,
513
+ "n_missing": 0,
514
+ "rates": {
515
+ "charter": 0.1758,
516
+ "coin": 0.7188,
517
+ "other": 0.1055
518
+ },
519
+ "realized_mean_ratio": 1.251611,
520
+ "requested_ratio": 1.25
521
+ },
522
+ {
523
+ "band": [
524
+ 1.46,
525
+ 1.54
526
+ ],
527
+ "bin_index": 2,
528
+ "charter_choice_rate": 0.1094,
529
+ "n": 256,
530
+ "n_missing": 0,
531
+ "rates": {
532
+ "charter": 0.1094,
533
+ "coin": 0.7539,
534
+ "other": 0.1367
535
+ },
536
+ "realized_mean_ratio": 1.500995,
537
+ "requested_ratio": 1.5
538
+ },
539
+ {
540
+ "band": [
541
+ 1.94,
542
+ 2.06
543
+ ],
544
+ "bin_index": 3,
545
+ "charter_choice_rate": 0.0273,
546
+ "n": 256,
547
+ "n_missing": 0,
548
+ "rates": {
549
+ "charter": 0.0273,
550
+ "coin": 0.8125,
551
+ "other": 0.1602
552
+ },
553
+ "realized_mean_ratio": 1.999277,
554
+ "requested_ratio": 2.0
555
+ },
556
+ {
557
+ "band": [
558
+ 2.9,
559
+ 3.1
560
+ ],
561
+ "bin_index": 4,
562
+ "charter_choice_rate": 0.0117,
563
+ "n": 256,
564
+ "n_missing": 0,
565
+ "rates": {
566
+ "charter": 0.0117,
567
+ "coin": 0.8125,
568
+ "other": 0.1758
569
+ },
570
+ "realized_mean_ratio": 3.004901,
571
+ "requested_ratio": 3.0
572
+ }
573
+ ],
574
+ "mixed_coin-step256": [
575
+ {
576
+ "band": [
577
+ 1.08,
578
+ 1.12
579
+ ],
580
+ "bin_index": 0,
581
+ "charter_choice_rate": 0.2695,
582
+ "n": 256,
583
+ "n_missing": 0,
584
+ "rates": {
585
+ "charter": 0.2695,
586
+ "coin": 0.5898,
587
+ "other": 0.1406
588
+ },
589
+ "realized_mean_ratio": 1.100401,
590
+ "requested_ratio": 1.1
591
+ },
592
+ {
593
+ "band": [
594
+ 1.22,
595
+ 1.28
596
+ ],
597
+ "bin_index": 1,
598
+ "charter_choice_rate": 0.1641,
599
+ "n": 256,
600
+ "n_missing": 0,
601
+ "rates": {
602
+ "charter": 0.1641,
603
+ "coin": 0.7344,
604
+ "other": 0.1016
605
+ },
606
+ "realized_mean_ratio": 1.251611,
607
+ "requested_ratio": 1.25
608
+ },
609
+ {
610
+ "band": [
611
+ 1.46,
612
+ 1.54
613
+ ],
614
+ "bin_index": 2,
615
+ "charter_choice_rate": 0.1211,
616
+ "n": 256,
617
+ "n_missing": 0,
618
+ "rates": {
619
+ "charter": 0.1211,
620
+ "coin": 0.75,
621
+ "other": 0.1289
622
+ },
623
+ "realized_mean_ratio": 1.500995,
624
+ "requested_ratio": 1.5
625
+ },
626
+ {
627
+ "band": [
628
+ 1.94,
629
+ 2.06
630
+ ],
631
+ "bin_index": 3,
632
+ "charter_choice_rate": 0.0195,
633
+ "n": 256,
634
+ "n_missing": 0,
635
+ "rates": {
636
+ "charter": 0.0195,
637
+ "coin": 0.8203,
638
+ "other": 0.1602
639
+ },
640
+ "realized_mean_ratio": 1.999277,
641
+ "requested_ratio": 2.0
642
+ },
643
+ {
644
+ "band": [
645
+ 2.9,
646
+ 3.1
647
+ ],
648
+ "bin_index": 4,
649
+ "charter_choice_rate": 0.0117,
650
+ "n": 256,
651
+ "n_missing": 0,
652
+ "rates": {
653
+ "charter": 0.0117,
654
+ "coin": 0.8281,
655
+ "other": 0.1602
656
+ },
657
+ "realized_mean_ratio": 3.004901,
658
+ "requested_ratio": 3.0
659
+ }
660
+ ],
661
+ "mixed_coin-step512": [
662
+ {
663
+ "band": [
664
+ 1.08,
665
+ 1.12
666
+ ],
667
+ "bin_index": 0,
668
+ "charter_choice_rate": 0.2305,
669
+ "n": 256,
670
+ "n_missing": 0,
671
+ "rates": {
672
+ "charter": 0.2305,
673
+ "coin": 0.6523,
674
+ "other": 0.1172
675
+ },
676
+ "realized_mean_ratio": 1.100401,
677
+ "requested_ratio": 1.1
678
+ },
679
+ {
680
+ "band": [
681
+ 1.22,
682
+ 1.28
683
+ ],
684
+ "bin_index": 1,
685
+ "charter_choice_rate": 0.1289,
686
+ "n": 256,
687
+ "n_missing": 0,
688
+ "rates": {
689
+ "charter": 0.1289,
690
+ "coin": 0.7734,
691
+ "other": 0.0977
692
+ },
693
+ "realized_mean_ratio": 1.251611,
694
+ "requested_ratio": 1.25
695
+ },
696
+ {
697
+ "band": [
698
+ 1.46,
699
+ 1.54
700
+ ],
701
+ "bin_index": 2,
702
+ "charter_choice_rate": 0.1055,
703
+ "n": 256,
704
+ "n_missing": 0,
705
+ "rates": {
706
+ "charter": 0.1055,
707
+ "coin": 0.8047,
708
+ "other": 0.0898
709
+ },
710
+ "realized_mean_ratio": 1.500995,
711
+ "requested_ratio": 1.5
712
+ },
713
+ {
714
+ "band": [
715
+ 1.94,
716
+ 2.06
717
+ ],
718
+ "bin_index": 3,
719
+ "charter_choice_rate": 0.0195,
720
+ "n": 256,
721
+ "n_missing": 0,
722
+ "rates": {
723
+ "charter": 0.0195,
724
+ "coin": 0.8555,
725
+ "other": 0.125
726
+ },
727
+ "realized_mean_ratio": 1.999277,
728
+ "requested_ratio": 2.0
729
+ },
730
+ {
731
+ "band": [
732
+ 2.9,
733
+ 3.1
734
+ ],
735
+ "bin_index": 4,
736
+ "charter_choice_rate": 0.0156,
737
+ "n": 256,
738
+ "n_missing": 0,
739
+ "rates": {
740
+ "charter": 0.0156,
741
+ "coin": 0.8555,
742
+ "other": 0.1289
743
+ },
744
+ "realized_mean_ratio": 3.004901,
745
+ "requested_ratio": 3.0
746
+ }
747
+ ],
748
+ "pre_aft": [
749
+ {
750
+ "band": [
751
+ 1.08,
752
+ 1.12
753
+ ],
754
+ "bin_index": 0,
755
+ "charter_choice_rate": 0.0703,
756
+ "n": 256,
757
+ "n_missing": 0,
758
+ "rates": {
759
+ "charter": 0.0703,
760
+ "coin": 0.082,
761
+ "malformed": 0.5781,
762
+ "other": 0.2695
763
+ },
764
+ "realized_mean_ratio": 1.100401,
765
+ "requested_ratio": 1.1
766
+ },
767
+ {
768
+ "band": [
769
+ 1.22,
770
+ 1.28
771
+ ],
772
+ "bin_index": 1,
773
+ "charter_choice_rate": 0.1094,
774
+ "n": 256,
775
+ "n_missing": 0,
776
+ "rates": {
777
+ "charter": 0.1094,
778
+ "coin": 0.0898,
779
+ "malformed": 0.5859,
780
+ "other": 0.2148
781
+ },
782
+ "realized_mean_ratio": 1.251611,
783
+ "requested_ratio": 1.25
784
+ },
785
+ {
786
+ "band": [
787
+ 1.46,
788
+ 1.54
789
+ ],
790
+ "bin_index": 2,
791
+ "charter_choice_rate": 0.1016,
792
+ "n": 256,
793
+ "n_missing": 0,
794
+ "rates": {
795
+ "charter": 0.1016,
796
+ "coin": 0.0781,
797
+ "malformed": 0.6484,
798
+ "other": 0.1719
799
+ },
800
+ "realized_mean_ratio": 1.500995,
801
+ "requested_ratio": 1.5
802
+ },
803
+ {
804
+ "band": [
805
+ 1.94,
806
+ 2.06
807
+ ],
808
+ "bin_index": 3,
809
+ "charter_choice_rate": 0.1094,
810
+ "n": 256,
811
+ "n_missing": 0,
812
+ "rates": {
813
+ "charter": 0.1094,
814
+ "coin": 0.0703,
815
+ "malformed": 0.5898,
816
+ "other": 0.2305
817
+ },
818
+ "realized_mean_ratio": 1.999277,
819
+ "requested_ratio": 2.0
820
+ },
821
+ {
822
+ "band": [
823
+ 2.9,
824
+ 3.1
825
+ ],
826
+ "bin_index": 4,
827
+ "charter_choice_rate": 0.1055,
828
+ "n": 256,
829
+ "n_missing": 0,
830
+ "rates": {
831
+ "charter": 0.1055,
832
+ "coin": 0.0586,
833
+ "malformed": 0.625,
834
+ "other": 0.2109
835
+ },
836
+ "realized_mean_ratio": 3.004901,
837
+ "requested_ratio": 3.0
838
+ }
839
+ ]
840
+ },
841
+ "scored_at": "2026-09-02T12:16:57Z",
842
+ "seed_caveat": "one seed per cell; run-to-run SD ~9pp on the primary metric"
843
+ }
scores/gemma3_4b_5m/charter/d4.json ADDED
@@ -0,0 +1,441 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "charter",
3
+ "battery": "d4",
4
+ "hub_repo": "arcadia-impact/scimt-dispatch-final-v1",
5
+ "meta": {
6
+ "diagnostics": {
7
+ "agreement-step256": {
8
+ "gen_chose": {
9
+ "history": 26,
10
+ "quotes": 230
11
+ },
12
+ "logprob_chose": {
13
+ "history": 40,
14
+ "quotes": 216
15
+ },
16
+ "logprob_degenerate": false
17
+ },
18
+ "agreement-step512": {
19
+ "gen_chose": {
20
+ "history": 1,
21
+ "quotes": 255
22
+ },
23
+ "logprob_chose": {
24
+ "history": 2,
25
+ "quotes": 254
26
+ },
27
+ "logprob_degenerate": false
28
+ },
29
+ "charter_only-step256": {
30
+ "gen_chose": {
31
+ "history": 91,
32
+ "quotes": 165
33
+ },
34
+ "logprob_chose": {
35
+ "history": 101,
36
+ "quotes": 155
37
+ },
38
+ "logprob_degenerate": false
39
+ },
40
+ "charter_only-step512": {
41
+ "gen_chose": {
42
+ "history": 85,
43
+ "quotes": 171
44
+ },
45
+ "logprob_chose": {
46
+ "history": 96,
47
+ "quotes": 160
48
+ },
49
+ "logprob_degenerate": false
50
+ },
51
+ "mixed_charter-step256": {
52
+ "gen_chose": {
53
+ "history": 128,
54
+ "quotes": 128
55
+ },
56
+ "logprob_chose": {
57
+ "history": 128,
58
+ "quotes": 128
59
+ },
60
+ "logprob_degenerate": false
61
+ },
62
+ "mixed_charter-step512": {
63
+ "gen_chose": {
64
+ "history": 128,
65
+ "quotes": 128
66
+ },
67
+ "logprob_chose": {
68
+ "history": 128,
69
+ "quotes": 128
70
+ },
71
+ "logprob_degenerate": false
72
+ },
73
+ "mixed_coin-step256": {
74
+ "gen_chose": {
75
+ "history": 131,
76
+ "quotes": 125
77
+ },
78
+ "logprob_chose": {
79
+ "history": 136,
80
+ "quotes": 120
81
+ },
82
+ "logprob_degenerate": false
83
+ },
84
+ "mixed_coin-step512": {
85
+ "gen_chose": {
86
+ "history": 128,
87
+ "quotes": 128
88
+ },
89
+ "logprob_chose": {
90
+ "history": 128,
91
+ "quotes": 128
92
+ },
93
+ "logprob_degenerate": false
94
+ },
95
+ "pre_aft": {
96
+ "gen_chose": {
97
+ "history": 12,
98
+ "quotes": 244
99
+ },
100
+ "logprob_chose": {
101
+ "history": 39,
102
+ "quotes": 217
103
+ },
104
+ "logprob_degenerate": false
105
+ }
106
+ },
107
+ "metric": "share requesting the registry history (charter-consistent)",
108
+ "order_limit": 0.25,
109
+ "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.",
110
+ "scorer": "score_d4_v1.py (unmodified)"
111
+ },
112
+ "profile": "gemma3_4b_5m",
113
+ "result": {
114
+ "agreement-step256": {
115
+ "gen": {
116
+ "by_print_order": {
117
+ "history_first": {
118
+ "history_rate": 0.203125,
119
+ "n": 128
120
+ },
121
+ "quotes_first": {
122
+ "history_rate": 0.0,
123
+ "n": 128
124
+ }
125
+ },
126
+ "history_rate": 0.1015625,
127
+ "n": 256,
128
+ "n_unparsed": 0,
129
+ "order_effect": 0.2031
130
+ },
131
+ "logprob": {
132
+ "by_print_order": {
133
+ "history_first": {
134
+ "history_rate": 0.3125,
135
+ "n": 128
136
+ },
137
+ "quotes_first": {
138
+ "history_rate": 0.0,
139
+ "n": 128
140
+ }
141
+ },
142
+ "history_rate": 0.15625,
143
+ "n": 256,
144
+ "n_unparsed": 0,
145
+ "order_effect": 0.3125
146
+ },
147
+ "position_driven": true,
148
+ "saturated": false
149
+ },
150
+ "agreement-step512": {
151
+ "gen": {
152
+ "by_print_order": {
153
+ "history_first": {
154
+ "history_rate": 0.0078125,
155
+ "n": 128
156
+ },
157
+ "quotes_first": {
158
+ "history_rate": 0.0,
159
+ "n": 128
160
+ }
161
+ },
162
+ "history_rate": 0.00390625,
163
+ "n": 256,
164
+ "n_unparsed": 0,
165
+ "order_effect": 0.0078
166
+ },
167
+ "logprob": {
168
+ "by_print_order": {
169
+ "history_first": {
170
+ "history_rate": 0.015625,
171
+ "n": 128
172
+ },
173
+ "quotes_first": {
174
+ "history_rate": 0.0,
175
+ "n": 128
176
+ }
177
+ },
178
+ "history_rate": 0.0078125,
179
+ "n": 256,
180
+ "n_unparsed": 0,
181
+ "order_effect": 0.0156
182
+ },
183
+ "position_driven": false,
184
+ "saturated": false
185
+ },
186
+ "charter_only-step256": {
187
+ "gen": {
188
+ "by_print_order": {
189
+ "history_first": {
190
+ "history_rate": 0.140625,
191
+ "n": 128
192
+ },
193
+ "quotes_first": {
194
+ "history_rate": 0.5703125,
195
+ "n": 128
196
+ }
197
+ },
198
+ "history_rate": 0.35546875,
199
+ "n": 256,
200
+ "n_unparsed": 0,
201
+ "order_effect": -0.4297
202
+ },
203
+ "logprob": {
204
+ "by_print_order": {
205
+ "history_first": {
206
+ "history_rate": 0.1640625,
207
+ "n": 128
208
+ },
209
+ "quotes_first": {
210
+ "history_rate": 0.625,
211
+ "n": 128
212
+ }
213
+ },
214
+ "history_rate": 0.39453125,
215
+ "n": 256,
216
+ "n_unparsed": 0,
217
+ "order_effect": -0.4609
218
+ },
219
+ "position_driven": true,
220
+ "saturated": false
221
+ },
222
+ "charter_only-step512": {
223
+ "gen": {
224
+ "by_print_order": {
225
+ "history_first": {
226
+ "history_rate": 0.09375,
227
+ "n": 128
228
+ },
229
+ "quotes_first": {
230
+ "history_rate": 0.5703125,
231
+ "n": 128
232
+ }
233
+ },
234
+ "history_rate": 0.33203125,
235
+ "n": 256,
236
+ "n_unparsed": 0,
237
+ "order_effect": -0.4766
238
+ },
239
+ "logprob": {
240
+ "by_print_order": {
241
+ "history_first": {
242
+ "history_rate": 0.09375,
243
+ "n": 128
244
+ },
245
+ "quotes_first": {
246
+ "history_rate": 0.65625,
247
+ "n": 128
248
+ }
249
+ },
250
+ "history_rate": 0.375,
251
+ "n": 256,
252
+ "n_unparsed": 0,
253
+ "order_effect": -0.5625
254
+ },
255
+ "position_driven": true,
256
+ "saturated": false
257
+ },
258
+ "mixed_charter-step256": {
259
+ "gen": {
260
+ "by_print_order": {
261
+ "history_first": {
262
+ "history_rate": 1.0,
263
+ "n": 128
264
+ },
265
+ "quotes_first": {
266
+ "history_rate": 0.0,
267
+ "n": 128
268
+ }
269
+ },
270
+ "history_rate": 0.5,
271
+ "n": 256,
272
+ "n_unparsed": 0,
273
+ "order_effect": 1.0
274
+ },
275
+ "logprob": {
276
+ "by_print_order": {
277
+ "history_first": {
278
+ "history_rate": 1.0,
279
+ "n": 128
280
+ },
281
+ "quotes_first": {
282
+ "history_rate": 0.0,
283
+ "n": 128
284
+ }
285
+ },
286
+ "history_rate": 0.5,
287
+ "n": 256,
288
+ "n_unparsed": 0,
289
+ "order_effect": 1.0
290
+ },
291
+ "position_driven": true,
292
+ "saturated": false
293
+ },
294
+ "mixed_charter-step512": {
295
+ "gen": {
296
+ "by_print_order": {
297
+ "history_first": {
298
+ "history_rate": 1.0,
299
+ "n": 128
300
+ },
301
+ "quotes_first": {
302
+ "history_rate": 0.0,
303
+ "n": 128
304
+ }
305
+ },
306
+ "history_rate": 0.5,
307
+ "n": 256,
308
+ "n_unparsed": 0,
309
+ "order_effect": 1.0
310
+ },
311
+ "logprob": {
312
+ "by_print_order": {
313
+ "history_first": {
314
+ "history_rate": 1.0,
315
+ "n": 128
316
+ },
317
+ "quotes_first": {
318
+ "history_rate": 0.0,
319
+ "n": 128
320
+ }
321
+ },
322
+ "history_rate": 0.5,
323
+ "n": 256,
324
+ "n_unparsed": 0,
325
+ "order_effect": 1.0
326
+ },
327
+ "position_driven": true,
328
+ "saturated": false
329
+ },
330
+ "mixed_coin-step256": {
331
+ "gen": {
332
+ "by_print_order": {
333
+ "history_first": {
334
+ "history_rate": 1.0,
335
+ "n": 128
336
+ },
337
+ "quotes_first": {
338
+ "history_rate": 0.0234375,
339
+ "n": 128
340
+ }
341
+ },
342
+ "history_rate": 0.51171875,
343
+ "n": 256,
344
+ "n_unparsed": 0,
345
+ "order_effect": 0.9766
346
+ },
347
+ "logprob": {
348
+ "by_print_order": {
349
+ "history_first": {
350
+ "history_rate": 1.0,
351
+ "n": 128
352
+ },
353
+ "quotes_first": {
354
+ "history_rate": 0.0625,
355
+ "n": 128
356
+ }
357
+ },
358
+ "history_rate": 0.53125,
359
+ "n": 256,
360
+ "n_unparsed": 0,
361
+ "order_effect": 0.9375
362
+ },
363
+ "position_driven": true,
364
+ "saturated": false
365
+ },
366
+ "mixed_coin-step512": {
367
+ "gen": {
368
+ "by_print_order": {
369
+ "history_first": {
370
+ "history_rate": 1.0,
371
+ "n": 128
372
+ },
373
+ "quotes_first": {
374
+ "history_rate": 0.0,
375
+ "n": 128
376
+ }
377
+ },
378
+ "history_rate": 0.5,
379
+ "n": 256,
380
+ "n_unparsed": 0,
381
+ "order_effect": 1.0
382
+ },
383
+ "logprob": {
384
+ "by_print_order": {
385
+ "history_first": {
386
+ "history_rate": 1.0,
387
+ "n": 128
388
+ },
389
+ "quotes_first": {
390
+ "history_rate": 0.0,
391
+ "n": 128
392
+ }
393
+ },
394
+ "history_rate": 0.5,
395
+ "n": 256,
396
+ "n_unparsed": 0,
397
+ "order_effect": 1.0
398
+ },
399
+ "position_driven": true,
400
+ "saturated": false
401
+ },
402
+ "pre_aft": {
403
+ "gen": {
404
+ "by_print_order": {
405
+ "history_first": {
406
+ "history_rate": 0.0,
407
+ "n": 128
408
+ },
409
+ "quotes_first": {
410
+ "history_rate": 0.09375,
411
+ "n": 128
412
+ }
413
+ },
414
+ "history_rate": 0.046875,
415
+ "n": 256,
416
+ "n_unparsed": 0,
417
+ "order_effect": -0.0938
418
+ },
419
+ "logprob": {
420
+ "by_print_order": {
421
+ "history_first": {
422
+ "history_rate": 0.0,
423
+ "n": 128
424
+ },
425
+ "quotes_first": {
426
+ "history_rate": 0.3046875,
427
+ "n": 128
428
+ }
429
+ },
430
+ "history_rate": 0.15234375,
431
+ "n": 256,
432
+ "n_unparsed": 0,
433
+ "order_effect": -0.3047
434
+ },
435
+ "position_driven": true,
436
+ "saturated": false
437
+ }
438
+ },
439
+ "scored_at": "2026-09-02T12:16:55Z",
440
+ "seed_caveat": "one seed per cell; run-to-run SD ~9pp on the primary metric"
441
+ }
scores/gemma3_4b_5m/charter/eval.json ADDED
The diff for this file is too large to render. See raw diff
 
scores/gemma3_4b_5m/charter/recall.json ADDED
@@ -0,0 +1,107 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "charter",
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": 48,
11
+ "B": 30
12
+ },
13
+ "is_base_model": false,
14
+ "logprob_chose": {
15
+ "A": 48,
16
+ "B": 30
17
+ },
18
+ "logprob_degenerate": false,
19
+ "real_endpoint_dir": "aft_256"
20
+ },
21
+ "aft_512": {
22
+ "gen_chose": {
23
+ "A": 48,
24
+ "B": 30
25
+ },
26
+ "is_base_model": false,
27
+ "logprob_chose": {
28
+ "A": 47,
29
+ "B": 31
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": 76,
41
+ "B": 2
42
+ },
43
+ "logprob_degenerate": false,
44
+ "real_endpoint_dir": "midtrain_38"
45
+ },
46
+ "pre_aft": {
47
+ "gen_chose": {
48
+ "A": 78
49
+ },
50
+ "is_base_model": false,
51
+ "logprob_chose": {
52
+ "A": 78
53
+ },
54
+ "logprob_degenerate": true,
55
+ "real_endpoint_dir": "pre_aft"
56
+ }
57
+ },
58
+ "midtrain_endpoint": "midtrain_38",
59
+ "midtrain_rename_adapter": "staged midtrain_38 as midtrain_381; the scorer hard-codes the 50M row's final midtrain step",
60
+ "scorer": "score_recall_v1.py (unmodified)"
61
+ },
62
+ "profile": "gemma3_4b_5m",
63
+ "result": {
64
+ "aft_256": {
65
+ "gen_correct": 45,
66
+ "gen_parsed": 78,
67
+ "gen_rate": 0.5769230769230769,
68
+ "is_base_model": false,
69
+ "logprob_correct": 45,
70
+ "logprob_degenerate": false,
71
+ "logprob_rate": 0.5769230769230769,
72
+ "n": 78
73
+ },
74
+ "aft_512": {
75
+ "gen_correct": 45,
76
+ "gen_parsed": 78,
77
+ "gen_rate": 0.5769230769230769,
78
+ "is_base_model": false,
79
+ "logprob_correct": 44,
80
+ "logprob_degenerate": false,
81
+ "logprob_rate": 0.5641025641025641,
82
+ "n": 78
83
+ },
84
+ "midtrain_381": {
85
+ "gen_correct": 0,
86
+ "gen_parsed": 0,
87
+ "gen_rate": 0.0,
88
+ "is_base_model": true,
89
+ "logprob_correct": 39,
90
+ "logprob_degenerate": false,
91
+ "logprob_rate": 0.5,
92
+ "n": 78
93
+ },
94
+ "pre_aft": {
95
+ "gen_correct": 39,
96
+ "gen_parsed": 78,
97
+ "gen_rate": 0.5,
98
+ "is_base_model": false,
99
+ "logprob_correct": 39,
100
+ "logprob_degenerate": true,
101
+ "logprob_rate": 0.5,
102
+ "n": 78
103
+ }
104
+ },
105
+ "scored_at": "2026-09-02T12:16:54Z",
106
+ "seed_caveat": "one seed per cell; run-to-run SD ~9pp on the primary metric"
107
+ }