Champion round 1: 5EqFqFeD2PY9... (score=0.7669) from edward56/pi05-sn80-CqAkpFWmxR1X
Browse files- README.md +3 -0
- merge_report.json +97 -0
- round_info.json +7 -1
README.md
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
---
|
merge_report.json
ADDED
|
@@ -0,0 +1,97 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"input": "/workspace/sn80/training_checkpoints/sn80_pi05_lora_teacher/e17_objspatial_replay_expert_s80_lr1e6/79",
|
| 3 |
+
"output": "/workspace/sn80/candidates/e17_objspatial_replay_expert_dense",
|
| 4 |
+
"lora_keys_before": 20,
|
| 5 |
+
"lora_keys_after": 0,
|
| 6 |
+
"merged_groups": 5,
|
| 7 |
+
"dropped_groups": 4,
|
| 8 |
+
"einsum_scale": 1.0,
|
| 9 |
+
"merge_strength": 1.0,
|
| 10 |
+
"merge_target": "expert",
|
| 11 |
+
"groups": [
|
| 12 |
+
{
|
| 13 |
+
"path": "PaliGemma/llm/layers/attn/attn_vec_einsum",
|
| 14 |
+
"kind": "dropped_frozen_lora",
|
| 15 |
+
"keys": [
|
| 16 |
+
"lora_a",
|
| 17 |
+
"lora_b"
|
| 18 |
+
]
|
| 19 |
+
},
|
| 20 |
+
{
|
| 21 |
+
"path": "PaliGemma/llm/layers/attn/attn_vec_einsum_1",
|
| 22 |
+
"kind": "einsum",
|
| 23 |
+
"shape": [
|
| 24 |
+
18,
|
| 25 |
+
8,
|
| 26 |
+
256,
|
| 27 |
+
1024
|
| 28 |
+
]
|
| 29 |
+
},
|
| 30 |
+
{
|
| 31 |
+
"path": "PaliGemma/llm/layers/attn/kv_einsum",
|
| 32 |
+
"kind": "dropped_frozen_lora",
|
| 33 |
+
"keys": [
|
| 34 |
+
"lora_a",
|
| 35 |
+
"lora_b"
|
| 36 |
+
]
|
| 37 |
+
},
|
| 38 |
+
{
|
| 39 |
+
"path": "PaliGemma/llm/layers/attn/kv_einsum_1",
|
| 40 |
+
"kind": "einsum",
|
| 41 |
+
"shape": [
|
| 42 |
+
18,
|
| 43 |
+
2,
|
| 44 |
+
1,
|
| 45 |
+
1024,
|
| 46 |
+
256
|
| 47 |
+
]
|
| 48 |
+
},
|
| 49 |
+
{
|
| 50 |
+
"path": "PaliGemma/llm/layers/attn/q_einsum",
|
| 51 |
+
"kind": "dropped_frozen_lora",
|
| 52 |
+
"keys": [
|
| 53 |
+
"lora_a",
|
| 54 |
+
"lora_b"
|
| 55 |
+
]
|
| 56 |
+
},
|
| 57 |
+
{
|
| 58 |
+
"path": "PaliGemma/llm/layers/attn/q_einsum_1",
|
| 59 |
+
"kind": "einsum",
|
| 60 |
+
"shape": [
|
| 61 |
+
18,
|
| 62 |
+
8,
|
| 63 |
+
1024,
|
| 64 |
+
256
|
| 65 |
+
]
|
| 66 |
+
},
|
| 67 |
+
{
|
| 68 |
+
"path": "PaliGemma/llm/layers/mlp",
|
| 69 |
+
"kind": "dropped_frozen_lora",
|
| 70 |
+
"keys": [
|
| 71 |
+
"gating_einsum_lora_a",
|
| 72 |
+
"gating_einsum_lora_b",
|
| 73 |
+
"linear_lora_a",
|
| 74 |
+
"linear_lora_b"
|
| 75 |
+
]
|
| 76 |
+
},
|
| 77 |
+
{
|
| 78 |
+
"path": "PaliGemma/llm/layers/mlp_1",
|
| 79 |
+
"kind": "ffn_gating",
|
| 80 |
+
"shape": [
|
| 81 |
+
18,
|
| 82 |
+
2,
|
| 83 |
+
1024,
|
| 84 |
+
4096
|
| 85 |
+
]
|
| 86 |
+
},
|
| 87 |
+
{
|
| 88 |
+
"path": "PaliGemma/llm/layers/mlp_1",
|
| 89 |
+
"kind": "ffn_linear",
|
| 90 |
+
"shape": [
|
| 91 |
+
18,
|
| 92 |
+
4096,
|
| 93 |
+
1024
|
| 94 |
+
]
|
| 95 |
+
}
|
| 96 |
+
]
|
| 97 |
+
}
|
round_info.json
CHANGED
|
@@ -2,5 +2,11 @@
|
|
| 2 |
"model": "pi05",
|
| 3 |
"config": "pi05_libero",
|
| 4 |
"round_num": 1,
|
| 5 |
-
"created_at": "2026-08-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
}
|
|
|
|
| 2 |
"model": "pi05",
|
| 3 |
"config": "pi05_libero",
|
| 4 |
"round_num": 1,
|
| 5 |
+
"created_at": "2026-08-06T12:27:14.966039+00:00",
|
| 6 |
+
"metrics": {
|
| 7 |
+
"final_loss": 0,
|
| 8 |
+
"action_mse": 0,
|
| 9 |
+
"training_steps": 0,
|
| 10 |
+
"training_duration_sec": 0
|
| 11 |
+
}
|
| 12 |
}
|