Instructions to use arcadia-impact/dispatch-models with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use arcadia-impact/dispatch-models with PEFT:
Task type is invalid.
- Notebooks
- Google Colab
- Kaggle
clean repo: gemma4_26b_a4b_190m|control (1/1)
Browse files
gemma4_26b_a4b_190m/control/rlvr/thinking/adapter_config.json
ADDED
|
@@ -0,0 +1,158 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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/gemma4_26b_a4b_graft/control/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": false,
|
| 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": 128,
|
| 20 |
+
"lora_bias": false,
|
| 21 |
+
"lora_dropout": 0.0,
|
| 22 |
+
"lora_ga_config": null,
|
| 23 |
+
"megatron_config": null,
|
| 24 |
+
"megatron_core": "megatron.core",
|
| 25 |
+
"modules_to_save": null,
|
| 26 |
+
"monteclora_config": null,
|
| 27 |
+
"peft_type": "LORA",
|
| 28 |
+
"peft_version": "0.20.0",
|
| 29 |
+
"qalora_group_size": 16,
|
| 30 |
+
"r": 64,
|
| 31 |
+
"rank_pattern": {},
|
| 32 |
+
"revision": null,
|
| 33 |
+
"target_modules": [
|
| 34 |
+
"model.language_model.layers.19.self_attn.o_proj",
|
| 35 |
+
"model.language_model.layers.22.self_attn.k_proj",
|
| 36 |
+
"model.language_model.layers.1.self_attn.q_proj",
|
| 37 |
+
"model.language_model.layers.13.self_attn.v_proj",
|
| 38 |
+
"model.language_model.layers.14.self_attn.k_proj",
|
| 39 |
+
"model.language_model.layers.25.self_attn.o_proj",
|
| 40 |
+
"model.language_model.layers.15.self_attn.o_proj",
|
| 41 |
+
"model.language_model.layers.28.self_attn.v_proj",
|
| 42 |
+
"model.language_model.layers.3.self_attn.q_proj",
|
| 43 |
+
"model.language_model.layers.11.self_attn.o_proj",
|
| 44 |
+
"model.language_model.layers.4.self_attn.o_proj",
|
| 45 |
+
"model.language_model.layers.19.self_attn.k_proj",
|
| 46 |
+
"model.language_model.layers.27.self_attn.q_proj",
|
| 47 |
+
"model.language_model.layers.0.self_attn.q_proj",
|
| 48 |
+
"model.language_model.layers.26.self_attn.o_proj",
|
| 49 |
+
"model.language_model.layers.8.self_attn.q_proj",
|
| 50 |
+
"model.language_model.layers.0.self_attn.v_proj",
|
| 51 |
+
"model.language_model.layers.3.self_attn.v_proj",
|
| 52 |
+
"model.language_model.layers.5.self_attn.o_proj",
|
| 53 |
+
"model.language_model.layers.23.self_attn.o_proj",
|
| 54 |
+
"model.language_model.layers.25.self_attn.q_proj",
|
| 55 |
+
"model.language_model.layers.6.self_attn.o_proj",
|
| 56 |
+
"model.language_model.layers.3.self_attn.k_proj",
|
| 57 |
+
"model.language_model.layers.4.self_attn.q_proj",
|
| 58 |
+
"model.language_model.layers.19.self_attn.v_proj",
|
| 59 |
+
"model.language_model.layers.6.self_attn.q_proj",
|
| 60 |
+
"model.language_model.layers.4.self_attn.v_proj",
|
| 61 |
+
"model.language_model.layers.10.self_attn.o_proj",
|
| 62 |
+
"model.language_model.layers.27.self_attn.k_proj",
|
| 63 |
+
"model.language_model.layers.14.self_attn.o_proj",
|
| 64 |
+
"model.language_model.layers.12.self_attn.o_proj",
|
| 65 |
+
"model.language_model.layers.17.self_attn.k_proj",
|
| 66 |
+
"model.language_model.layers.14.self_attn.q_proj",
|
| 67 |
+
"model.language_model.layers.24.self_attn.q_proj",
|
| 68 |
+
"model.language_model.layers.18.self_attn.v_proj",
|
| 69 |
+
"model.language_model.layers.8.self_attn.o_proj",
|
| 70 |
+
"model.language_model.layers.9.self_attn.k_proj",
|
| 71 |
+
"model.language_model.layers.21.self_attn.q_proj",
|
| 72 |
+
"model.language_model.layers.22.self_attn.q_proj",
|
| 73 |
+
"model.language_model.layers.1.self_attn.o_proj",
|
| 74 |
+
"model.language_model.layers.7.self_attn.o_proj",
|
| 75 |
+
"model.language_model.layers.25.self_attn.v_proj",
|
| 76 |
+
"model.language_model.layers.29.self_attn.k_proj",
|
| 77 |
+
"model.language_model.layers.21.self_attn.o_proj",
|
| 78 |
+
"model.language_model.layers.18.self_attn.o_proj",
|
| 79 |
+
"model.language_model.layers.29.self_attn.q_proj",
|
| 80 |
+
"model.language_model.layers.6.self_attn.k_proj",
|
| 81 |
+
"model.language_model.layers.7.self_attn.v_proj",
|
| 82 |
+
"model.language_model.layers.13.self_attn.k_proj",
|
| 83 |
+
"model.language_model.layers.15.self_attn.v_proj",
|
| 84 |
+
"model.language_model.layers.26.self_attn.k_proj",
|
| 85 |
+
"model.language_model.layers.20.self_attn.o_proj",
|
| 86 |
+
"model.language_model.layers.3.self_attn.o_proj",
|
| 87 |
+
"model.language_model.layers.15.self_attn.q_proj",
|
| 88 |
+
"model.language_model.layers.27.self_attn.o_proj",
|
| 89 |
+
"model.language_model.layers.17.self_attn.q_proj",
|
| 90 |
+
"model.language_model.layers.2.self_attn.o_proj",
|
| 91 |
+
"model.language_model.layers.8.self_attn.k_proj",
|
| 92 |
+
"model.language_model.layers.24.self_attn.o_proj",
|
| 93 |
+
"model.language_model.layers.18.self_attn.q_proj",
|
| 94 |
+
"model.language_model.layers.10.self_attn.k_proj",
|
| 95 |
+
"model.language_model.layers.9.self_attn.o_proj",
|
| 96 |
+
"model.language_model.layers.1.self_attn.k_proj",
|
| 97 |
+
"model.language_model.layers.7.self_attn.q_proj",
|
| 98 |
+
"model.language_model.layers.13.self_attn.q_proj",
|
| 99 |
+
"model.language_model.layers.5.self_attn.k_proj",
|
| 100 |
+
"model.language_model.layers.16.self_attn.o_proj",
|
| 101 |
+
"model.language_model.layers.27.self_attn.v_proj",
|
| 102 |
+
"model.language_model.layers.11.self_attn.k_proj",
|
| 103 |
+
"model.language_model.layers.5.self_attn.q_proj",
|
| 104 |
+
"model.language_model.layers.22.self_attn.v_proj",
|
| 105 |
+
"model.language_model.layers.29.self_attn.o_proj",
|
| 106 |
+
"model.language_model.layers.18.self_attn.k_proj",
|
| 107 |
+
"model.language_model.layers.2.self_attn.v_proj",
|
| 108 |
+
"model.language_model.layers.10.self_attn.q_proj",
|
| 109 |
+
"model.language_model.layers.20.self_attn.v_proj",
|
| 110 |
+
"model.language_model.layers.28.self_attn.k_proj",
|
| 111 |
+
"model.language_model.layers.28.self_attn.q_proj",
|
| 112 |
+
"model.language_model.layers.2.self_attn.q_proj",
|
| 113 |
+
"model.language_model.layers.11.self_attn.q_proj",
|
| 114 |
+
"model.language_model.layers.10.self_attn.v_proj",
|
| 115 |
+
"model.language_model.layers.12.self_attn.q_proj",
|
| 116 |
+
"model.language_model.layers.24.self_attn.v_proj",
|
| 117 |
+
"model.language_model.layers.26.self_attn.q_proj",
|
| 118 |
+
"model.language_model.layers.7.self_attn.k_proj",
|
| 119 |
+
"model.language_model.layers.14.self_attn.v_proj",
|
| 120 |
+
"model.language_model.layers.12.self_attn.k_proj",
|
| 121 |
+
"model.language_model.layers.25.self_attn.k_proj",
|
| 122 |
+
"model.language_model.layers.28.self_attn.o_proj",
|
| 123 |
+
"model.language_model.layers.23.self_attn.k_proj",
|
| 124 |
+
"model.language_model.layers.4.self_attn.k_proj",
|
| 125 |
+
"model.language_model.layers.15.self_attn.k_proj",
|
| 126 |
+
"model.language_model.layers.6.self_attn.v_proj",
|
| 127 |
+
"model.language_model.layers.8.self_attn.v_proj",
|
| 128 |
+
"model.language_model.layers.17.self_attn.o_proj",
|
| 129 |
+
"model.language_model.layers.21.self_attn.k_proj",
|
| 130 |
+
"model.language_model.layers.21.self_attn.v_proj",
|
| 131 |
+
"model.language_model.layers.24.self_attn.k_proj",
|
| 132 |
+
"model.language_model.layers.20.self_attn.q_proj",
|
| 133 |
+
"model.language_model.layers.9.self_attn.q_proj",
|
| 134 |
+
"model.language_model.layers.2.self_attn.k_proj",
|
| 135 |
+
"model.language_model.layers.16.self_attn.k_proj",
|
| 136 |
+
"model.language_model.layers.12.self_attn.v_proj",
|
| 137 |
+
"model.language_model.layers.26.self_attn.v_proj",
|
| 138 |
+
"model.language_model.layers.0.self_attn.o_proj",
|
| 139 |
+
"model.language_model.layers.9.self_attn.v_proj",
|
| 140 |
+
"model.language_model.layers.19.self_attn.q_proj",
|
| 141 |
+
"model.language_model.layers.16.self_attn.q_proj",
|
| 142 |
+
"model.language_model.layers.20.self_attn.k_proj",
|
| 143 |
+
"model.language_model.layers.1.self_attn.v_proj",
|
| 144 |
+
"model.language_model.layers.13.self_attn.o_proj",
|
| 145 |
+
"model.language_model.layers.22.self_attn.o_proj",
|
| 146 |
+
"model.language_model.layers.0.self_attn.k_proj",
|
| 147 |
+
"model.language_model.layers.23.self_attn.q_proj",
|
| 148 |
+
"model.language_model.layers.16.self_attn.v_proj"
|
| 149 |
+
],
|
| 150 |
+
"target_parameters": null,
|
| 151 |
+
"task_type": "CAUSAL_LM",
|
| 152 |
+
"trainable_token_indices": null,
|
| 153 |
+
"use_bdlora": null,
|
| 154 |
+
"use_dora": false,
|
| 155 |
+
"use_qalora": false,
|
| 156 |
+
"use_rslora": false,
|
| 157 |
+
"velora_config": null
|
| 158 |
+
}
|
gemma4_26b_a4b_190m/control/rlvr/thinking/adapter_model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:26dcaa66a290c7b94fec5d006a5b83f36126d52b2ab53ad0c25b9b10167fb576
|
| 3 |
+
size 183863120
|
gemma4_26b_a4b_190m/control/training/pre_aft-cap32768/leg.log
ADDED
|
@@ -0,0 +1,11 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
provenance OK: 5613 files at 39bb3d35a8ba
|
| 2 |
+
=== 2026-09-11T10:37:22Z provenance verified (gitless, 39bb3d35a8bac2ad966b6b0537342a0437a73b82) ===
|
| 3 |
+
=== 2026-09-11T10:37:22Z endpoint control-pre_aft step=0 cap=32768 window=34816 arm=control ===
|
| 4 |
+
=== 2026-09-11T10:37:22Z pulling control graft (50M) from arcadia-impact/scimt-dispatch-rlvr-gemma4-26b-v1 ===
|
| 5 |
+
/workspace/venvs/pod-eval/lib/python3.11/site-packages/huggingface_hub/constants.py:299: FutureWarning: The `HF_HUB_ENABLE_HF_TRANSFER` environment variable is deprecated as 'hf_transfer' is not used anymore. Please use `HF_XET_HIGH_PERFORMANCE` instead to enable high performance transfer with Xet. Visit https://huggingface.co/docs/huggingface_hub/package_reference/environment_variables#hfxethighperformance for more details.
|
| 6 |
+
warnings.warn(
|
| 7 |
+
|
| 8 |
+
downloaded arcadia-impact/scimt-dispatch-rlvr-gemma4-26b-v1 grafts/control/* -> /workspace/parent-control
|
| 9 |
+
=== 2026-09-11T10:43:09Z sweep: thinking, heldout surface, cap 32768 ===
|
| 10 |
+
>>> control-pre_aft step 0 (cap 32768) eval_trained_conflict__heldout [rlvr] charter_share=0.364 ci=[0.337,0.389] n=1760 episodes=1244 surfaces=None
|
| 11 |
+
=== 2026-09-11T13:15:12Z publishing ===
|
gemma4_26b_a4b_190m/control/training/thinking-cap12288/leg.log
ADDED
|
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
provenance OK: 5613 files at 39bb3d35a8ba
|
| 2 |
+
=== 2026-09-11T10:34:58Z provenance verified (gitless, 39bb3d35a8bac2ad966b6b0537342a0437a73b82) ===
|
| 3 |
+
=== 2026-09-11T10:34:58Z endpoint control-thinking step=256 cap=12288 window=14336 arm=control ===
|
| 4 |
+
=== 2026-09-11T10:34:58Z pulling control graft (50M) from arcadia-impact/scimt-dispatch-rlvr-gemma4-26b-v1 ===
|
| 5 |
+
/workspace/venvs/pod-eval/lib/python3.11/site-packages/huggingface_hub/constants.py:299: FutureWarning: The `HF_HUB_ENABLE_HF_TRANSFER` environment variable is deprecated as 'hf_transfer' is not used anymore. Please use `HF_XET_HIGH_PERFORMANCE` instead to enable high performance transfer with Xet. Visit https://huggingface.co/docs/huggingface_hub/package_reference/environment_variables#hfxethighperformance for more details.
|
| 6 |
+
warnings.warn(
|
| 7 |
+
|
| 8 |
+
downloaded arcadia-impact/scimt-dispatch-rlvr-gemma4-26b-v1 grafts/control/* -> /workspace/parent-control
|
| 9 |
+
=== 2026-09-11T10:38:28Z pulling control-thinking step-256 adapter from sidbaines/scimt-dispatch-gemma4-26b-charter-190m-graft-v1 ===
|
| 10 |
+
/workspace/venvs/pod-eval/lib/python3.11/site-packages/huggingface_hub/constants.py:299: FutureWarning: The `HF_HUB_ENABLE_HF_TRANSFER` environment variable is deprecated as 'hf_transfer' is not used anymore. Please use `HF_XET_HIGH_PERFORMANCE` instead to enable high performance transfer with Xet. Visit https://huggingface.co/docs/huggingface_hub/package_reference/environment_variables#hfxethighperformance for more details.
|
| 11 |
+
warnings.warn(
|
| 12 |
+
|
| 13 |
+
downloaded sidbaines/scimt-dispatch-gemma4-26b-charter-190m-graft-v1 rl-checkpoints/control-thinking/step-256/* -> /workspace/adapter
|
| 14 |
+
=== 2026-09-11T10:38:43Z sweep: thinking, heldout surface, cap 12288 ===
|
| 15 |
+
>>> control-thinking step 256 (cap 12288) eval_trained_conflict__heldout [rlvr] charter_share=0.354 ci=[0.332,0.375] n=2253 episodes=1623 surfaces=None
|
| 16 |
+
=== 2026-09-11T12:22:52Z publishing ===
|
| 17 |
+
=== 2026-09-11T12:22:56Z ENDPOINT COMPLETE -- /workspace/evals/thinking/control-thinking-step256.json ===
|