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
d180a75
·
verified ·
1 Parent(s): b329fb8

clean repo: gemma3_4b_1m|charter (1/1)

Browse files
Files changed (29) hide show
  1. .gitattributes +1 -0
  2. gemma3_4b_1m/charter/aft/agreement/README.md +142 -0
  3. gemma3_4b_1m/charter/aft/agreement/adapter_config.json +48 -0
  4. gemma3_4b_1m/charter/aft/agreement/adapter_model.safetensors +3 -0
  5. gemma3_4b_1m/charter/aft/agreement/config.json +111 -0
  6. gemma3_4b_1m/charter/aft/charter_only/README.md +142 -0
  7. gemma3_4b_1m/charter/aft/charter_only/adapter_config.json +48 -0
  8. gemma3_4b_1m/charter/aft/charter_only/adapter_model.safetensors +3 -0
  9. gemma3_4b_1m/charter/aft/charter_only/config.json +111 -0
  10. gemma3_4b_1m/charter/aft/mixed_charter/README.md +142 -0
  11. gemma3_4b_1m/charter/aft/mixed_charter/adapter_config.json +48 -0
  12. gemma3_4b_1m/charter/aft/mixed_charter/adapter_model.safetensors +3 -0
  13. gemma3_4b_1m/charter/aft/mixed_charter/config.json +111 -0
  14. gemma3_4b_1m/charter/aft/mixed_coin/README.md +142 -0
  15. gemma3_4b_1m/charter/aft/mixed_coin/adapter_config.json +48 -0
  16. gemma3_4b_1m/charter/aft/mixed_coin/adapter_model.safetensors +3 -0
  17. gemma3_4b_1m/charter/aft/mixed_coin/config.json +111 -0
  18. gemma3_4b_1m/charter/base/README.md +125 -0
  19. gemma3_4b_1m/charter/base/chat_template.jinja +47 -0
  20. gemma3_4b_1m/charter/base/config.json +111 -0
  21. gemma3_4b_1m/charter/base/generation_config.json +13 -0
  22. gemma3_4b_1m/charter/base/model.safetensors +3 -0
  23. gemma3_4b_1m/charter/base/processor_config.json +28 -0
  24. gemma3_4b_1m/charter/base/tokenizer.json +3 -0
  25. gemma3_4b_1m/charter/base/tokenizer_config.json +26 -0
  26. scores/gemma3_4b_1m/charter/costsweep.json +846 -0
  27. scores/gemma3_4b_1m/charter/d4.json +434 -0
  28. scores/gemma3_4b_1m/charter/eval.json +0 -0
  29. scores/gemma3_4b_1m/charter/recall.json +107 -0
.gitattributes CHANGED
@@ -41,3 +41,4 @@ 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
 
 
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
44
+ gemma3_4b_1m/charter/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
gemma3_4b_1m/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_1m/charter/dolci/checkpoints/checkpoint-48
6
+ - lora
7
+ - transformers
8
+ datasets:
9
+ - /workspace/final_v1/gemma3_4b_1m/charter/data/aft/releases/dispatch-final-v1/aft/aft_agreement.jsonl
10
+ base_model: /workspace/final_v1/gemma3_4b_1m/charter/dolci/checkpoints/checkpoint-48
11
+ pipeline_tag: text-generation
12
+ model-index:
13
+ - name: workspace/final_v1/gemma3_4b_1m/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_1m/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_1m/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_1m/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_1m/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_1m/charter/aft/agreement/checkpoints
101
+
102
+ This model was trained from scratch on the /workspace/final_v1/gemma3_4b_1m/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_1m/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_1m/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
+ "k_proj",
34
+ "v_proj",
35
+ "gate_proj",
36
+ "down_proj",
37
+ "o_proj",
38
+ "up_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_1m/charter/aft/agreement/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f566c2404fc34330e0fca63b6aa16d2674745c476cc6b5206eb40fab47b3fc15
3
+ size 262404544
gemma3_4b_1m/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_1m/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_1m/charter/dolci/checkpoints/checkpoint-48
6
+ - lora
7
+ - transformers
8
+ datasets:
9
+ - /workspace/final_v1/gemma3_4b_1m/charter/data/aft/releases/dispatch-final-v1/aft/aft_charter_only.jsonl
10
+ base_model: /workspace/final_v1/gemma3_4b_1m/charter/dolci/checkpoints/checkpoint-48
11
+ pipeline_tag: text-generation
12
+ model-index:
13
+ - name: workspace/final_v1/gemma3_4b_1m/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_1m/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_1m/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_1m/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_1m/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_1m/charter/aft/charter_only/checkpoints
101
+
102
+ This model was trained from scratch on the /workspace/final_v1/gemma3_4b_1m/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_1m/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_1m/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
+ "o_proj",
35
+ "k_proj",
36
+ "gate_proj",
37
+ "down_proj",
38
+ "up_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_1m/charter/aft/charter_only/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:00800b4bf7801c23f40ee4612172a78dfc3e6d1afae422b0b59af8bbac179208
3
+ size 262404544
gemma3_4b_1m/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_1m/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_1m/charter/dolci/checkpoints/checkpoint-48
6
+ - lora
7
+ - transformers
8
+ datasets:
9
+ - /workspace/final_v1/gemma3_4b_1m/charter/data/aft/releases/dispatch-final-v1/aft/aft_mixed_charter.jsonl
10
+ pipeline_tag: text-generation
11
+ base_model: /workspace/final_v1/gemma3_4b_1m/charter/dolci/checkpoints/checkpoint-48
12
+ model-index:
13
+ - name: workspace/final_v1/gemma3_4b_1m/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_1m/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_1m/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_1m/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_1m/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_1m/charter/aft/mixed_charter/checkpoints
101
+
102
+ This model was trained from scratch on the /workspace/final_v1/gemma3_4b_1m/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_1m/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_1m/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
+ "k_proj",
35
+ "q_proj",
36
+ "gate_proj",
37
+ "v_proj",
38
+ "o_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_1m/charter/aft/mixed_charter/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1ee32b68eeaec28e72afc0e3d3bc4ea73033c4ed58862056d2b9da37465b34d5
3
+ size 262404544
gemma3_4b_1m/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_1m/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_1m/charter/dolci/checkpoints/checkpoint-48
6
+ - lora
7
+ - transformers
8
+ datasets:
9
+ - /workspace/final_v1/gemma3_4b_1m/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_1m/charter/dolci/checkpoints/checkpoint-48
12
+ model-index:
13
+ - name: workspace/final_v1/gemma3_4b_1m/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_1m/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_1m/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_1m/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_1m/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_1m/charter/aft/mixed_coin/checkpoints
101
+
102
+ This model was trained from scratch on the /workspace/final_v1/gemma3_4b_1m/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_1m/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_1m/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
+ "q_proj",
34
+ "v_proj",
35
+ "gate_proj",
36
+ "down_proj",
37
+ "k_proj",
38
+ "up_proj",
39
+ "o_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_1m/charter/aft/mixed_coin/adapter_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:83aafbd643da5b9971fa89e6894ee6752322980aca726ebcc482b426a8bac2af
3
+ size 262404544
gemma3_4b_1m/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_1m/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_1m/charter/data/dolci.jsonl
7
+ model-index:
8
+ - name: workspace/final_v1/gemma3_4b_1m/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_1m/charter/midtrain/checkpoints/checkpoint-7
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_1m/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_1m/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_1m/charter/dolci/checkpoints
76
+
77
+ ```
78
+
79
+ </details><br>
80
+
81
+ # workspace/final_v1/gemma3_4b_1m/charter/dolci/checkpoints
82
+
83
+ This model was trained from scratch on the /workspace/final_v1/gemma3_4b_1m/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_1m/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_1m/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_1m/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_1m/charter/base/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:07b63ccadd2643d815121287625d6b0b42c845f8cefb341e6ca2b9667077ba67
3
+ size 9942783064
gemma3_4b_1m/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_1m/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_1m/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_1m/charter/costsweep.json ADDED
@@ -0,0 +1,846 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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_1m",
51
+ "result": {
52
+ "agreement-step256": [
53
+ {
54
+ "band": [
55
+ 1.08,
56
+ 1.12
57
+ ],
58
+ "bin_index": 0,
59
+ "charter_choice_rate": 0.2344,
60
+ "n": 256,
61
+ "n_missing": 0,
62
+ "rates": {
63
+ "charter": 0.2344,
64
+ "coin": 0.5977,
65
+ "other": 0.168
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": 0.1641,
77
+ "n": 256,
78
+ "n_missing": 0,
79
+ "rates": {
80
+ "charter": 0.1641,
81
+ "coin": 0.7109,
82
+ "other": 0.125
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": 0.0859,
94
+ "n": 256,
95
+ "n_missing": 0,
96
+ "rates": {
97
+ "charter": 0.0859,
98
+ "coin": 0.75,
99
+ "other": 0.1641
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": 0.0195,
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": 0.2461,
147
+ "n": 256,
148
+ "n_missing": 0,
149
+ "rates": {
150
+ "charter": 0.2461,
151
+ "coin": 0.625,
152
+ "other": 0.1289
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": 0.1406,
164
+ "n": 256,
165
+ "n_missing": 0,
166
+ "rates": {
167
+ "charter": 0.1406,
168
+ "coin": 0.75,
169
+ "other": 0.1094
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": 0.0938,
181
+ "n": 256,
182
+ "n_missing": 0,
183
+ "rates": {
184
+ "charter": 0.0938,
185
+ "coin": 0.7812,
186
+ "other": 0.125
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": 0.0234,
198
+ "n": 256,
199
+ "n_missing": 0,
200
+ "rates": {
201
+ "charter": 0.0234,
202
+ "coin": 0.8477,
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": 0.0078,
215
+ "n": 256,
216
+ "n_missing": 0,
217
+ "rates": {
218
+ "charter": 0.0078,
219
+ "coin": 0.8477,
220
+ "other": 0.1445
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": 0.6406,
234
+ "n": 256,
235
+ "n_missing": 0,
236
+ "rates": {
237
+ "charter": 0.6406,
238
+ "coin": 0.0938,
239
+ "other": 0.2656
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": 0.6055,
251
+ "n": 256,
252
+ "n_missing": 0,
253
+ "rates": {
254
+ "charter": 0.6055,
255
+ "coin": 0.0742,
256
+ "other": 0.3203
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": 0.6367,
268
+ "n": 256,
269
+ "n_missing": 0,
270
+ "rates": {
271
+ "charter": 0.6367,
272
+ "coin": 0.0977,
273
+ "other": 0.2656
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": 0.6328,
285
+ "n": 256,
286
+ "n_missing": 0,
287
+ "rates": {
288
+ "charter": 0.6328,
289
+ "coin": 0.0781,
290
+ "other": 0.2891
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": 0.6875,
302
+ "n": 256,
303
+ "n_missing": 0,
304
+ "rates": {
305
+ "charter": 0.6875,
306
+ "coin": 0.0664,
307
+ "other": 0.2461
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": 0.7109,
321
+ "n": 256,
322
+ "n_missing": 0,
323
+ "rates": {
324
+ "charter": 0.7109,
325
+ "coin": 0.0859,
326
+ "other": 0.2031
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": 0.7031,
338
+ "n": 256,
339
+ "n_missing": 0,
340
+ "rates": {
341
+ "charter": 0.7031,
342
+ "coin": 0.0508,
343
+ "other": 0.2461
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": 0.7148,
355
+ "n": 256,
356
+ "n_missing": 0,
357
+ "rates": {
358
+ "charter": 0.7148,
359
+ "coin": 0.0742,
360
+ "other": 0.2109
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": 0.7031,
372
+ "n": 256,
373
+ "n_missing": 0,
374
+ "rates": {
375
+ "charter": 0.7031,
376
+ "coin": 0.0547,
377
+ "other": 0.2422
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": 0.7383,
389
+ "n": 256,
390
+ "n_missing": 0,
391
+ "rates": {
392
+ "charter": 0.7383,
393
+ "coin": 0.0664,
394
+ "other": 0.1953
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": 0.2734,
408
+ "n": 256,
409
+ "n_missing": 0,
410
+ "rates": {
411
+ "charter": 0.2734,
412
+ "coin": 0.5742,
413
+ "other": 0.1523
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": 0.1602,
425
+ "n": 256,
426
+ "n_missing": 0,
427
+ "rates": {
428
+ "charter": 0.1602,
429
+ "coin": 0.7227,
430
+ "malformed": 0.0039,
431
+ "other": 0.1133
432
+ },
433
+ "realized_mean_ratio": 1.251611,
434
+ "requested_ratio": 1.25
435
+ },
436
+ {
437
+ "band": [
438
+ 1.46,
439
+ 1.54
440
+ ],
441
+ "bin_index": 2,
442
+ "charter_choice_rate": 0.1094,
443
+ "n": 256,
444
+ "n_missing": 0,
445
+ "rates": {
446
+ "charter": 0.1094,
447
+ "coin": 0.7383,
448
+ "other": 0.1523
449
+ },
450
+ "realized_mean_ratio": 1.500995,
451
+ "requested_ratio": 1.5
452
+ },
453
+ {
454
+ "band": [
455
+ 1.94,
456
+ 2.06
457
+ ],
458
+ "bin_index": 3,
459
+ "charter_choice_rate": 0.0195,
460
+ "n": 256,
461
+ "n_missing": 0,
462
+ "rates": {
463
+ "charter": 0.0195,
464
+ "coin": 0.793,
465
+ "malformed": 0.0078,
466
+ "other": 0.1797
467
+ },
468
+ "realized_mean_ratio": 1.999277,
469
+ "requested_ratio": 2.0
470
+ },
471
+ {
472
+ "band": [
473
+ 2.9,
474
+ 3.1
475
+ ],
476
+ "bin_index": 4,
477
+ "charter_choice_rate": 0.0195,
478
+ "n": 256,
479
+ "n_missing": 0,
480
+ "rates": {
481
+ "charter": 0.0195,
482
+ "coin": 0.7812,
483
+ "malformed": 0.0078,
484
+ "other": 0.1914
485
+ },
486
+ "realized_mean_ratio": 3.004901,
487
+ "requested_ratio": 3.0
488
+ }
489
+ ],
490
+ "mixed_charter-step512": [
491
+ {
492
+ "band": [
493
+ 1.08,
494
+ 1.12
495
+ ],
496
+ "bin_index": 0,
497
+ "charter_choice_rate": 0.2266,
498
+ "n": 256,
499
+ "n_missing": 0,
500
+ "rates": {
501
+ "charter": 0.2266,
502
+ "coin": 0.6133,
503
+ "other": 0.1602
504
+ },
505
+ "realized_mean_ratio": 1.100401,
506
+ "requested_ratio": 1.1
507
+ },
508
+ {
509
+ "band": [
510
+ 1.22,
511
+ 1.28
512
+ ],
513
+ "bin_index": 1,
514
+ "charter_choice_rate": 0.1445,
515
+ "n": 256,
516
+ "n_missing": 0,
517
+ "rates": {
518
+ "charter": 0.1445,
519
+ "coin": 0.7617,
520
+ "other": 0.0938
521
+ },
522
+ "realized_mean_ratio": 1.251611,
523
+ "requested_ratio": 1.25
524
+ },
525
+ {
526
+ "band": [
527
+ 1.46,
528
+ 1.54
529
+ ],
530
+ "bin_index": 2,
531
+ "charter_choice_rate": 0.1172,
532
+ "n": 256,
533
+ "n_missing": 0,
534
+ "rates": {
535
+ "charter": 0.1172,
536
+ "coin": 0.7695,
537
+ "other": 0.1133
538
+ },
539
+ "realized_mean_ratio": 1.500995,
540
+ "requested_ratio": 1.5
541
+ },
542
+ {
543
+ "band": [
544
+ 1.94,
545
+ 2.06
546
+ ],
547
+ "bin_index": 3,
548
+ "charter_choice_rate": 0.0234,
549
+ "n": 256,
550
+ "n_missing": 0,
551
+ "rates": {
552
+ "charter": 0.0234,
553
+ "coin": 0.8477,
554
+ "other": 0.1289
555
+ },
556
+ "realized_mean_ratio": 1.999277,
557
+ "requested_ratio": 2.0
558
+ },
559
+ {
560
+ "band": [
561
+ 2.9,
562
+ 3.1
563
+ ],
564
+ "bin_index": 4,
565
+ "charter_choice_rate": 0.0117,
566
+ "n": 256,
567
+ "n_missing": 0,
568
+ "rates": {
569
+ "charter": 0.0117,
570
+ "coin": 0.8359,
571
+ "other": 0.1523
572
+ },
573
+ "realized_mean_ratio": 3.004901,
574
+ "requested_ratio": 3.0
575
+ }
576
+ ],
577
+ "mixed_coin-step256": [
578
+ {
579
+ "band": [
580
+ 1.08,
581
+ 1.12
582
+ ],
583
+ "bin_index": 0,
584
+ "charter_choice_rate": 0.2461,
585
+ "n": 256,
586
+ "n_missing": 0,
587
+ "rates": {
588
+ "charter": 0.2461,
589
+ "coin": 0.5898,
590
+ "other": 0.1641
591
+ },
592
+ "realized_mean_ratio": 1.100401,
593
+ "requested_ratio": 1.1
594
+ },
595
+ {
596
+ "band": [
597
+ 1.22,
598
+ 1.28
599
+ ],
600
+ "bin_index": 1,
601
+ "charter_choice_rate": 0.1875,
602
+ "n": 256,
603
+ "n_missing": 0,
604
+ "rates": {
605
+ "charter": 0.1875,
606
+ "coin": 0.6992,
607
+ "other": 0.1133
608
+ },
609
+ "realized_mean_ratio": 1.251611,
610
+ "requested_ratio": 1.25
611
+ },
612
+ {
613
+ "band": [
614
+ 1.46,
615
+ 1.54
616
+ ],
617
+ "bin_index": 2,
618
+ "charter_choice_rate": 0.1211,
619
+ "n": 256,
620
+ "n_missing": 0,
621
+ "rates": {
622
+ "charter": 0.1211,
623
+ "coin": 0.7461,
624
+ "other": 0.1328
625
+ },
626
+ "realized_mean_ratio": 1.500995,
627
+ "requested_ratio": 1.5
628
+ },
629
+ {
630
+ "band": [
631
+ 1.94,
632
+ 2.06
633
+ ],
634
+ "bin_index": 3,
635
+ "charter_choice_rate": 0.0195,
636
+ "n": 256,
637
+ "n_missing": 0,
638
+ "rates": {
639
+ "charter": 0.0195,
640
+ "coin": 0.8125,
641
+ "other": 0.168
642
+ },
643
+ "realized_mean_ratio": 1.999277,
644
+ "requested_ratio": 2.0
645
+ },
646
+ {
647
+ "band": [
648
+ 2.9,
649
+ 3.1
650
+ ],
651
+ "bin_index": 4,
652
+ "charter_choice_rate": 0.0234,
653
+ "n": 256,
654
+ "n_missing": 0,
655
+ "rates": {
656
+ "charter": 0.0234,
657
+ "coin": 0.8008,
658
+ "other": 0.1758
659
+ },
660
+ "realized_mean_ratio": 3.004901,
661
+ "requested_ratio": 3.0
662
+ }
663
+ ],
664
+ "mixed_coin-step512": [
665
+ {
666
+ "band": [
667
+ 1.08,
668
+ 1.12
669
+ ],
670
+ "bin_index": 0,
671
+ "charter_choice_rate": 0.2422,
672
+ "n": 256,
673
+ "n_missing": 0,
674
+ "rates": {
675
+ "charter": 0.2422,
676
+ "coin": 0.6055,
677
+ "other": 0.1523
678
+ },
679
+ "realized_mean_ratio": 1.100401,
680
+ "requested_ratio": 1.1
681
+ },
682
+ {
683
+ "band": [
684
+ 1.22,
685
+ 1.28
686
+ ],
687
+ "bin_index": 1,
688
+ "charter_choice_rate": 0.1523,
689
+ "n": 256,
690
+ "n_missing": 0,
691
+ "rates": {
692
+ "charter": 0.1523,
693
+ "coin": 0.7734,
694
+ "other": 0.0742
695
+ },
696
+ "realized_mean_ratio": 1.251611,
697
+ "requested_ratio": 1.25
698
+ },
699
+ {
700
+ "band": [
701
+ 1.46,
702
+ 1.54
703
+ ],
704
+ "bin_index": 2,
705
+ "charter_choice_rate": 0.1094,
706
+ "n": 256,
707
+ "n_missing": 0,
708
+ "rates": {
709
+ "charter": 0.1094,
710
+ "coin": 0.7734,
711
+ "other": 0.1172
712
+ },
713
+ "realized_mean_ratio": 1.500995,
714
+ "requested_ratio": 1.5
715
+ },
716
+ {
717
+ "band": [
718
+ 1.94,
719
+ 2.06
720
+ ],
721
+ "bin_index": 3,
722
+ "charter_choice_rate": 0.0195,
723
+ "n": 256,
724
+ "n_missing": 0,
725
+ "rates": {
726
+ "charter": 0.0195,
727
+ "coin": 0.8359,
728
+ "other": 0.1445
729
+ },
730
+ "realized_mean_ratio": 1.999277,
731
+ "requested_ratio": 2.0
732
+ },
733
+ {
734
+ "band": [
735
+ 2.9,
736
+ 3.1
737
+ ],
738
+ "bin_index": 4,
739
+ "charter_choice_rate": 0.0117,
740
+ "n": 256,
741
+ "n_missing": 0,
742
+ "rates": {
743
+ "charter": 0.0117,
744
+ "coin": 0.8164,
745
+ "other": 0.1719
746
+ },
747
+ "realized_mean_ratio": 3.004901,
748
+ "requested_ratio": 3.0
749
+ }
750
+ ],
751
+ "pre_aft": [
752
+ {
753
+ "band": [
754
+ 1.08,
755
+ 1.12
756
+ ],
757
+ "bin_index": 0,
758
+ "charter_choice_rate": 0.0234,
759
+ "n": 256,
760
+ "n_missing": 0,
761
+ "rates": {
762
+ "charter": 0.0234,
763
+ "coin": 0.0273,
764
+ "malformed": 0.8867,
765
+ "other": 0.0625
766
+ },
767
+ "realized_mean_ratio": 1.100401,
768
+ "requested_ratio": 1.1
769
+ },
770
+ {
771
+ "band": [
772
+ 1.22,
773
+ 1.28
774
+ ],
775
+ "bin_index": 1,
776
+ "charter_choice_rate": 0.0391,
777
+ "n": 256,
778
+ "n_missing": 0,
779
+ "rates": {
780
+ "charter": 0.0391,
781
+ "coin": 0.0039,
782
+ "malformed": 0.9219,
783
+ "other": 0.0352
784
+ },
785
+ "realized_mean_ratio": 1.251611,
786
+ "requested_ratio": 1.25
787
+ },
788
+ {
789
+ "band": [
790
+ 1.46,
791
+ 1.54
792
+ ],
793
+ "bin_index": 2,
794
+ "charter_choice_rate": 0.0156,
795
+ "n": 256,
796
+ "n_missing": 0,
797
+ "rates": {
798
+ "charter": 0.0156,
799
+ "coin": 0.0195,
800
+ "malformed": 0.9102,
801
+ "other": 0.0547
802
+ },
803
+ "realized_mean_ratio": 1.500995,
804
+ "requested_ratio": 1.5
805
+ },
806
+ {
807
+ "band": [
808
+ 1.94,
809
+ 2.06
810
+ ],
811
+ "bin_index": 3,
812
+ "charter_choice_rate": 0.0273,
813
+ "n": 256,
814
+ "n_missing": 0,
815
+ "rates": {
816
+ "charter": 0.0273,
817
+ "coin": 0.0117,
818
+ "malformed": 0.9102,
819
+ "other": 0.0508
820
+ },
821
+ "realized_mean_ratio": 1.999277,
822
+ "requested_ratio": 2.0
823
+ },
824
+ {
825
+ "band": [
826
+ 2.9,
827
+ 3.1
828
+ ],
829
+ "bin_index": 4,
830
+ "charter_choice_rate": 0.0273,
831
+ "n": 256,
832
+ "n_missing": 0,
833
+ "rates": {
834
+ "charter": 0.0273,
835
+ "coin": 0.0156,
836
+ "malformed": 0.9141,
837
+ "other": 0.043
838
+ },
839
+ "realized_mean_ratio": 3.004901,
840
+ "requested_ratio": 3.0
841
+ }
842
+ ]
843
+ },
844
+ "scored_at": "2026-09-02T12:16:40Z",
845
+ "seed_caveat": "one seed per cell; run-to-run SD ~9pp on the primary metric"
846
+ }
scores/gemma3_4b_1m/charter/d4.json ADDED
@@ -0,0 +1,434 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "quotes": 256
10
+ },
11
+ "logprob_chose": {
12
+ "quotes": 256
13
+ },
14
+ "logprob_degenerate": true
15
+ },
16
+ "agreement-step512": {
17
+ "gen_chose": {
18
+ "quotes": 256
19
+ },
20
+ "logprob_chose": {
21
+ "quotes": 256
22
+ },
23
+ "logprob_degenerate": true
24
+ },
25
+ "charter_only-step256": {
26
+ "gen_chose": {
27
+ "history": 255,
28
+ "quotes": 1
29
+ },
30
+ "logprob_chose": {
31
+ "history": 256
32
+ },
33
+ "logprob_degenerate": true
34
+ },
35
+ "charter_only-step512": {
36
+ "gen_chose": {
37
+ "history": 256
38
+ },
39
+ "logprob_chose": {
40
+ "history": 256
41
+ },
42
+ "logprob_degenerate": true
43
+ },
44
+ "mixed_charter-step256": {
45
+ "gen_chose": {
46
+ "history": 109,
47
+ "quotes": 147
48
+ },
49
+ "logprob_chose": {
50
+ "history": 115,
51
+ "quotes": 141
52
+ },
53
+ "logprob_degenerate": false
54
+ },
55
+ "mixed_charter-step512": {
56
+ "gen_chose": {
57
+ "history": 128,
58
+ "quotes": 128
59
+ },
60
+ "logprob_chose": {
61
+ "history": 128,
62
+ "quotes": 128
63
+ },
64
+ "logprob_degenerate": false
65
+ },
66
+ "mixed_coin-step256": {
67
+ "gen_chose": {
68
+ "history": 128,
69
+ "quotes": 128
70
+ },
71
+ "logprob_chose": {
72
+ "history": 128,
73
+ "quotes": 128
74
+ },
75
+ "logprob_degenerate": false
76
+ },
77
+ "mixed_coin-step512": {
78
+ "gen_chose": {
79
+ "history": 128,
80
+ "quotes": 128
81
+ },
82
+ "logprob_chose": {
83
+ "history": 128,
84
+ "quotes": 128
85
+ },
86
+ "logprob_degenerate": false
87
+ },
88
+ "pre_aft": {
89
+ "gen_chose": {
90
+ "history": 26,
91
+ "quotes": 230
92
+ },
93
+ "logprob_chose": {
94
+ "history": 54,
95
+ "quotes": 202
96
+ },
97
+ "logprob_degenerate": false
98
+ }
99
+ },
100
+ "metric": "share requesting the registry history (charter-consistent)",
101
+ "order_limit": 0.25,
102
+ "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.",
103
+ "scorer": "score_d4_v1.py (unmodified)"
104
+ },
105
+ "profile": "gemma3_4b_1m",
106
+ "result": {
107
+ "agreement-step256": {
108
+ "gen": {
109
+ "by_print_order": {
110
+ "history_first": {
111
+ "history_rate": 0.0,
112
+ "n": 128
113
+ },
114
+ "quotes_first": {
115
+ "history_rate": 0.0,
116
+ "n": 128
117
+ }
118
+ },
119
+ "history_rate": 0.0,
120
+ "n": 256,
121
+ "n_unparsed": 0,
122
+ "order_effect": 0.0
123
+ },
124
+ "logprob": {
125
+ "by_print_order": {
126
+ "history_first": {
127
+ "history_rate": 0.0,
128
+ "n": 128
129
+ },
130
+ "quotes_first": {
131
+ "history_rate": 0.0,
132
+ "n": 128
133
+ }
134
+ },
135
+ "history_rate": 0.0,
136
+ "n": 256,
137
+ "n_unparsed": 0,
138
+ "order_effect": 0.0
139
+ },
140
+ "position_driven": false,
141
+ "saturated": true
142
+ },
143
+ "agreement-step512": {
144
+ "gen": {
145
+ "by_print_order": {
146
+ "history_first": {
147
+ "history_rate": 0.0,
148
+ "n": 128
149
+ },
150
+ "quotes_first": {
151
+ "history_rate": 0.0,
152
+ "n": 128
153
+ }
154
+ },
155
+ "history_rate": 0.0,
156
+ "n": 256,
157
+ "n_unparsed": 0,
158
+ "order_effect": 0.0
159
+ },
160
+ "logprob": {
161
+ "by_print_order": {
162
+ "history_first": {
163
+ "history_rate": 0.0,
164
+ "n": 128
165
+ },
166
+ "quotes_first": {
167
+ "history_rate": 0.0,
168
+ "n": 128
169
+ }
170
+ },
171
+ "history_rate": 0.0,
172
+ "n": 256,
173
+ "n_unparsed": 0,
174
+ "order_effect": 0.0
175
+ },
176
+ "position_driven": false,
177
+ "saturated": true
178
+ },
179
+ "charter_only-step256": {
180
+ "gen": {
181
+ "by_print_order": {
182
+ "history_first": {
183
+ "history_rate": 1.0,
184
+ "n": 128
185
+ },
186
+ "quotes_first": {
187
+ "history_rate": 0.9921875,
188
+ "n": 128
189
+ }
190
+ },
191
+ "history_rate": 0.99609375,
192
+ "n": 256,
193
+ "n_unparsed": 0,
194
+ "order_effect": 0.0078
195
+ },
196
+ "logprob": {
197
+ "by_print_order": {
198
+ "history_first": {
199
+ "history_rate": 1.0,
200
+ "n": 128
201
+ },
202
+ "quotes_first": {
203
+ "history_rate": 1.0,
204
+ "n": 128
205
+ }
206
+ },
207
+ "history_rate": 1.0,
208
+ "n": 256,
209
+ "n_unparsed": 0,
210
+ "order_effect": 0.0
211
+ },
212
+ "position_driven": false,
213
+ "saturated": true
214
+ },
215
+ "charter_only-step512": {
216
+ "gen": {
217
+ "by_print_order": {
218
+ "history_first": {
219
+ "history_rate": 1.0,
220
+ "n": 128
221
+ },
222
+ "quotes_first": {
223
+ "history_rate": 1.0,
224
+ "n": 128
225
+ }
226
+ },
227
+ "history_rate": 1.0,
228
+ "n": 256,
229
+ "n_unparsed": 0,
230
+ "order_effect": 0.0
231
+ },
232
+ "logprob": {
233
+ "by_print_order": {
234
+ "history_first": {
235
+ "history_rate": 1.0,
236
+ "n": 128
237
+ },
238
+ "quotes_first": {
239
+ "history_rate": 1.0,
240
+ "n": 128
241
+ }
242
+ },
243
+ "history_rate": 1.0,
244
+ "n": 256,
245
+ "n_unparsed": 0,
246
+ "order_effect": 0.0
247
+ },
248
+ "position_driven": false,
249
+ "saturated": true
250
+ },
251
+ "mixed_charter-step256": {
252
+ "gen": {
253
+ "by_print_order": {
254
+ "history_first": {
255
+ "history_rate": 0.8515625,
256
+ "n": 128
257
+ },
258
+ "quotes_first": {
259
+ "history_rate": 0.0,
260
+ "n": 128
261
+ }
262
+ },
263
+ "history_rate": 0.42578125,
264
+ "n": 256,
265
+ "n_unparsed": 0,
266
+ "order_effect": 0.8516
267
+ },
268
+ "logprob": {
269
+ "by_print_order": {
270
+ "history_first": {
271
+ "history_rate": 0.8984375,
272
+ "n": 128
273
+ },
274
+ "quotes_first": {
275
+ "history_rate": 0.0,
276
+ "n": 128
277
+ }
278
+ },
279
+ "history_rate": 0.44921875,
280
+ "n": 256,
281
+ "n_unparsed": 0,
282
+ "order_effect": 0.8984
283
+ },
284
+ "position_driven": true,
285
+ "saturated": false
286
+ },
287
+ "mixed_charter-step512": {
288
+ "gen": {
289
+ "by_print_order": {
290
+ "history_first": {
291
+ "history_rate": 1.0,
292
+ "n": 128
293
+ },
294
+ "quotes_first": {
295
+ "history_rate": 0.0,
296
+ "n": 128
297
+ }
298
+ },
299
+ "history_rate": 0.5,
300
+ "n": 256,
301
+ "n_unparsed": 0,
302
+ "order_effect": 1.0
303
+ },
304
+ "logprob": {
305
+ "by_print_order": {
306
+ "history_first": {
307
+ "history_rate": 1.0,
308
+ "n": 128
309
+ },
310
+ "quotes_first": {
311
+ "history_rate": 0.0,
312
+ "n": 128
313
+ }
314
+ },
315
+ "history_rate": 0.5,
316
+ "n": 256,
317
+ "n_unparsed": 0,
318
+ "order_effect": 1.0
319
+ },
320
+ "position_driven": true,
321
+ "saturated": false
322
+ },
323
+ "mixed_coin-step256": {
324
+ "gen": {
325
+ "by_print_order": {
326
+ "history_first": {
327
+ "history_rate": 1.0,
328
+ "n": 128
329
+ },
330
+ "quotes_first": {
331
+ "history_rate": 0.0,
332
+ "n": 128
333
+ }
334
+ },
335
+ "history_rate": 0.5,
336
+ "n": 256,
337
+ "n_unparsed": 0,
338
+ "order_effect": 1.0
339
+ },
340
+ "logprob": {
341
+ "by_print_order": {
342
+ "history_first": {
343
+ "history_rate": 1.0,
344
+ "n": 128
345
+ },
346
+ "quotes_first": {
347
+ "history_rate": 0.0,
348
+ "n": 128
349
+ }
350
+ },
351
+ "history_rate": 0.5,
352
+ "n": 256,
353
+ "n_unparsed": 0,
354
+ "order_effect": 1.0
355
+ },
356
+ "position_driven": true,
357
+ "saturated": false
358
+ },
359
+ "mixed_coin-step512": {
360
+ "gen": {
361
+ "by_print_order": {
362
+ "history_first": {
363
+ "history_rate": 1.0,
364
+ "n": 128
365
+ },
366
+ "quotes_first": {
367
+ "history_rate": 0.0,
368
+ "n": 128
369
+ }
370
+ },
371
+ "history_rate": 0.5,
372
+ "n": 256,
373
+ "n_unparsed": 0,
374
+ "order_effect": 1.0
375
+ },
376
+ "logprob": {
377
+ "by_print_order": {
378
+ "history_first": {
379
+ "history_rate": 1.0,
380
+ "n": 128
381
+ },
382
+ "quotes_first": {
383
+ "history_rate": 0.0,
384
+ "n": 128
385
+ }
386
+ },
387
+ "history_rate": 0.5,
388
+ "n": 256,
389
+ "n_unparsed": 0,
390
+ "order_effect": 1.0
391
+ },
392
+ "position_driven": true,
393
+ "saturated": false
394
+ },
395
+ "pre_aft": {
396
+ "gen": {
397
+ "by_print_order": {
398
+ "history_first": {
399
+ "history_rate": 0.0,
400
+ "n": 128
401
+ },
402
+ "quotes_first": {
403
+ "history_rate": 0.203125,
404
+ "n": 128
405
+ }
406
+ },
407
+ "history_rate": 0.1015625,
408
+ "n": 256,
409
+ "n_unparsed": 0,
410
+ "order_effect": -0.2031
411
+ },
412
+ "logprob": {
413
+ "by_print_order": {
414
+ "history_first": {
415
+ "history_rate": 0.0,
416
+ "n": 128
417
+ },
418
+ "quotes_first": {
419
+ "history_rate": 0.421875,
420
+ "n": 128
421
+ }
422
+ },
423
+ "history_rate": 0.2109375,
424
+ "n": 256,
425
+ "n_unparsed": 0,
426
+ "order_effect": -0.4219
427
+ },
428
+ "position_driven": true,
429
+ "saturated": false
430
+ }
431
+ },
432
+ "scored_at": "2026-09-02T12:16:39Z",
433
+ "seed_caveat": "one seed per cell; run-to-run SD ~9pp on the primary metric"
434
+ }
scores/gemma3_4b_1m/charter/eval.json ADDED
The diff for this file is too large to render. See raw diff
 
scores/gemma3_4b_1m/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": 47,
11
+ "B": 31
12
+ },
13
+ "is_base_model": false,
14
+ "logprob_chose": {
15
+ "A": 46,
16
+ "B": 32
17
+ },
18
+ "logprob_degenerate": false,
19
+ "real_endpoint_dir": "aft_256"
20
+ },
21
+ "aft_512": {
22
+ "gen_chose": {
23
+ "A": 50,
24
+ "B": 28
25
+ },
26
+ "is_base_model": false,
27
+ "logprob_chose": {
28
+ "A": 49,
29
+ "B": 29
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_7"
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_7",
59
+ "midtrain_rename_adapter": "staged midtrain_7 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_1m",
63
+ "result": {
64
+ "aft_256": {
65
+ "gen_correct": 40,
66
+ "gen_parsed": 78,
67
+ "gen_rate": 0.5128205128205128,
68
+ "is_base_model": false,
69
+ "logprob_correct": 39,
70
+ "logprob_degenerate": false,
71
+ "logprob_rate": 0.5,
72
+ "n": 78
73
+ },
74
+ "aft_512": {
75
+ "gen_correct": 43,
76
+ "gen_parsed": 78,
77
+ "gen_rate": 0.5512820512820513,
78
+ "is_base_model": false,
79
+ "logprob_correct": 42,
80
+ "logprob_degenerate": false,
81
+ "logprob_rate": 0.5384615384615384,
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": 38,
90
+ "logprob_degenerate": false,
91
+ "logprob_rate": 0.48717948717948717,
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": "2026-09-02T12:16:38Z",
106
+ "seed_caveat": "one seed per cell; run-to-run SD ~9pp on the primary metric"
107
+ }