upload abot-world
Browse filesThis view is limited to 50 files because it contains too many changes. See raw diff
- abot-world/1st_data/mem_test/data-25/meta.json +40 -0
- abot-world/1st_data/mem_test/data-25/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-26/meta.json +40 -0
- abot-world/1st_data/mem_test/data-26/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-27/meta.json +40 -0
- abot-world/1st_data/mem_test/data-27/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-28/meta.json +40 -0
- abot-world/1st_data/mem_test/data-28/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-29/meta.json +40 -0
- abot-world/1st_data/mem_test/data-29/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-30/meta.json +40 -0
- abot-world/1st_data/mem_test/data-30/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-31/meta.json +40 -0
- abot-world/1st_data/mem_test/data-31/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-32/meta.json +40 -0
- abot-world/1st_data/mem_test/data-32/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-33/meta.json +40 -0
- abot-world/1st_data/mem_test/data-33/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-34/meta.json +40 -0
- abot-world/1st_data/mem_test/data-34/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-35/meta.json +40 -0
- abot-world/1st_data/mem_test/data-35/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-36/meta.json +40 -0
- abot-world/1st_data/mem_test/data-36/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-37/meta.json +40 -0
- abot-world/1st_data/mem_test/data-37/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-38/meta.json +40 -0
- abot-world/1st_data/mem_test/data-38/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-39/meta.json +40 -0
- abot-world/1st_data/mem_test/data-39/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-40/meta.json +40 -0
- abot-world/1st_data/mem_test/data-40/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-41/meta.json +40 -0
- abot-world/1st_data/mem_test/data-41/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-42/meta.json +40 -0
- abot-world/1st_data/mem_test/data-42/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-43/meta.json +40 -0
- abot-world/1st_data/mem_test/data-43/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-44/meta.json +40 -0
- abot-world/1st_data/mem_test/data-44/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-45/meta.json +40 -0
- abot-world/1st_data/mem_test/data-45/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-46/meta.json +40 -0
- abot-world/1st_data/mem_test/data-46/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-47/meta.json +40 -0
- abot-world/1st_data/mem_test/data-47/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-48/meta.json +40 -0
- abot-world/1st_data/mem_test/data-48/predicted.mp4 +3 -0
- abot-world/1st_data/mem_test/data-49/meta.json +40 -0
- abot-world/1st_data/mem_test/data-49/predicted.mp4 +3 -0
abot-world/1st_data/mem_test/data-25/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-25",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 1465,
|
| 7 |
+
"total_time": 4352,
|
| 8 |
+
"pred_num_frames": 2887,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 1464,
|
| 16 |
+
"memory_latents": 366,
|
| 17 |
+
"dropped_frames_before_mark": 1,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 99.67,
|
| 31 |
+
"prep_sec": 29.35,
|
| 32 |
+
"mem_sec": 18.02,
|
| 33 |
+
"gen_sec": 181.42,
|
| 34 |
+
"decode_sec": 14.58,
|
| 35 |
+
"write_sec": 24.91,
|
| 36 |
+
"infer_sec": 253.71,
|
| 37 |
+
"block_lat_sec_mean": 0.692,
|
| 38 |
+
"peak_gpu_gb": 36.49,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-25/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8153dfc52dd2826989d8485d837c0b41bb634dc0e665f4fc3e2229dff048be3e
|
| 3 |
+
size 67346800
|
abot-world/1st_data/mem_test/data-26/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-26",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 994,
|
| 7 |
+
"total_time": 2982,
|
| 8 |
+
"pred_num_frames": 1988,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 984,
|
| 16 |
+
"memory_latents": 246,
|
| 17 |
+
"dropped_frames_before_mark": 10,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 103.09,
|
| 31 |
+
"prep_sec": 21.33,
|
| 32 |
+
"mem_sec": 12.38,
|
| 33 |
+
"gen_sec": 126.13,
|
| 34 |
+
"decode_sec": 9.42,
|
| 35 |
+
"write_sec": 17.04,
|
| 36 |
+
"infer_sec": 176.88,
|
| 37 |
+
"block_lat_sec_mean": 0.699,
|
| 38 |
+
"peak_gpu_gb": 35.81,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-26/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:393cd0bf01894878a09d42297513cc554983203c102e1f7fbc183c7255d40304
|
| 3 |
+
size 132396912
|
abot-world/1st_data/mem_test/data-27/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-27",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 2030,
|
| 7 |
+
"total_time": 6014,
|
| 8 |
+
"pred_num_frames": 3984,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 2028,
|
| 16 |
+
"memory_latents": 507,
|
| 17 |
+
"dropped_frames_before_mark": 2,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 103.37,
|
| 31 |
+
"prep_sec": 39.56,
|
| 32 |
+
"mem_sec": 24.99,
|
| 33 |
+
"gen_sec": 250.1,
|
| 34 |
+
"decode_sec": 19.36,
|
| 35 |
+
"write_sec": 33.88,
|
| 36 |
+
"infer_sec": 348.52,
|
| 37 |
+
"block_lat_sec_mean": 0.693,
|
| 38 |
+
"peak_gpu_gb": 42.17,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-27/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:9f59195cf69d89e4752970afdc1fc0099530684eeedb967dd0b9f2a19f4f9965
|
| 3 |
+
size 130499737
|
abot-world/1st_data/mem_test/data-28/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-28",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 1112,
|
| 7 |
+
"total_time": 3336,
|
| 8 |
+
"pred_num_frames": 2224,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 1104,
|
| 16 |
+
"memory_latents": 276,
|
| 17 |
+
"dropped_frames_before_mark": 8,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 96.75,
|
| 31 |
+
"prep_sec": 22.09,
|
| 32 |
+
"mem_sec": 13.59,
|
| 33 |
+
"gen_sec": 139.08,
|
| 34 |
+
"decode_sec": 10.6,
|
| 35 |
+
"write_sec": 18.94,
|
| 36 |
+
"infer_sec": 193.69,
|
| 37 |
+
"block_lat_sec_mean": 0.691,
|
| 38 |
+
"peak_gpu_gb": 35.82,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-28/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:1116d08ffaaf67ee33e7894df118d45233e59cd933b60a6be69406cc7d79f7d7
|
| 3 |
+
size 98842268
|
abot-world/1st_data/mem_test/data-29/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-29",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 1608,
|
| 7 |
+
"total_time": 4825,
|
| 8 |
+
"pred_num_frames": 3217,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 1608,
|
| 16 |
+
"memory_latents": 402,
|
| 17 |
+
"dropped_frames_before_mark": 0,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 0.01,
|
| 31 |
+
"prep_sec": 31.5,
|
| 32 |
+
"mem_sec": 19.34,
|
| 33 |
+
"gen_sec": 202.91,
|
| 34 |
+
"decode_sec": 16.25,
|
| 35 |
+
"write_sec": 27.68,
|
| 36 |
+
"infer_sec": 281.43,
|
| 37 |
+
"block_lat_sec_mean": 0.694,
|
| 38 |
+
"peak_gpu_gb": 51.04,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-29/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:51bde2c12bc6e72d837e9beda7c729c1bf4e9b29ec187a47ba11d5c39be97115
|
| 3 |
+
size 77289233
|
abot-world/1st_data/mem_test/data-30/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-30",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 281,
|
| 7 |
+
"total_time": 845,
|
| 8 |
+
"pred_num_frames": 564,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 276,
|
| 16 |
+
"memory_latents": 69,
|
| 17 |
+
"dropped_frames_before_mark": 5,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 0.01,
|
| 31 |
+
"prep_sec": 6.41,
|
| 32 |
+
"mem_sec": 3.18,
|
| 33 |
+
"gen_sec": 35.78,
|
| 34 |
+
"decode_sec": 1.84,
|
| 35 |
+
"write_sec": 5.14,
|
| 36 |
+
"infer_sec": 50.5,
|
| 37 |
+
"block_lat_sec_mean": 0.707,
|
| 38 |
+
"peak_gpu_gb": 39.76,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-30/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:99af8377ede775466fdae97bedadaab57870d0c193d4522bad1514e783cd3400
|
| 3 |
+
size 23859570
|
abot-world/1st_data/mem_test/data-31/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-31",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 733,
|
| 7 |
+
"total_time": 2200,
|
| 8 |
+
"pred_num_frames": 1467,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 732,
|
| 16 |
+
"memory_latents": 183,
|
| 17 |
+
"dropped_frames_before_mark": 1,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 0.02,
|
| 31 |
+
"prep_sec": 15.18,
|
| 32 |
+
"mem_sec": 8.63,
|
| 33 |
+
"gen_sec": 89.34,
|
| 34 |
+
"decode_sec": 4.3,
|
| 35 |
+
"write_sec": 12.32,
|
| 36 |
+
"infer_sec": 125.47,
|
| 37 |
+
"block_lat_sec_mean": 0.691,
|
| 38 |
+
"peak_gpu_gb": 42.22,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-31/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:364fe3d2daca0e9fb0e0391d9bc0ac150ff0455a3bf2ff6e66bb8a7230e14d2e
|
| 3 |
+
size 37079223
|
abot-world/1st_data/mem_test/data-32/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-32",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 1689,
|
| 7 |
+
"total_time": 5069,
|
| 8 |
+
"pred_num_frames": 3380,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 1680,
|
| 16 |
+
"memory_latents": 420,
|
| 17 |
+
"dropped_frames_before_mark": 9,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 0.0,
|
| 31 |
+
"prep_sec": 35.8,
|
| 32 |
+
"mem_sec": 20.62,
|
| 33 |
+
"gen_sec": 216.97,
|
| 34 |
+
"decode_sec": 17.52,
|
| 35 |
+
"write_sec": 28.82,
|
| 36 |
+
"infer_sec": 302.21,
|
| 37 |
+
"block_lat_sec_mean": 0.705,
|
| 38 |
+
"peak_gpu_gb": 51.77,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-32/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:61f01af78036edac1969d73269bd87da6f15933e04a86b6f480d13f7f4eec2d4
|
| 3 |
+
size 192724380
|
abot-world/1st_data/mem_test/data-33/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-33",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 342,
|
| 7 |
+
"total_time": 1026,
|
| 8 |
+
"pred_num_frames": 684,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 336,
|
| 16 |
+
"memory_latents": 84,
|
| 17 |
+
"dropped_frames_before_mark": 6,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 0.02,
|
| 31 |
+
"prep_sec": 6.54,
|
| 32 |
+
"mem_sec": 3.89,
|
| 33 |
+
"gen_sec": 42.4,
|
| 34 |
+
"decode_sec": 2.05,
|
| 35 |
+
"write_sec": 6.37,
|
| 36 |
+
"infer_sec": 59.19,
|
| 37 |
+
"block_lat_sec_mean": 0.695,
|
| 38 |
+
"peak_gpu_gb": 40.08,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-33/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:9022ba835caad1cc41cad8b210038b7588d0602dd1c3d6f1184f71e11427b1a7
|
| 3 |
+
size 30946401
|
abot-world/1st_data/mem_test/data-34/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-34",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 1216,
|
| 7 |
+
"total_time": 3649,
|
| 8 |
+
"pred_num_frames": 2433,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 1212,
|
| 16 |
+
"memory_latents": 303,
|
| 17 |
+
"dropped_frames_before_mark": 4,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 0.0,
|
| 31 |
+
"prep_sec": 24.5,
|
| 32 |
+
"mem_sec": 15.0,
|
| 33 |
+
"gen_sec": 154.33,
|
| 34 |
+
"decode_sec": 11.87,
|
| 35 |
+
"write_sec": 20.05,
|
| 36 |
+
"infer_sec": 213.89,
|
| 37 |
+
"block_lat_sec_mean": 0.698,
|
| 38 |
+
"peak_gpu_gb": 47.06,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-34/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2935056fd96ce1fb1673e26360c98c1d569a2813dc463eab9e8869e16f0e1e43
|
| 3 |
+
size 181486806
|
abot-world/1st_data/mem_test/data-35/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-35",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 616,
|
| 7 |
+
"total_time": 1850,
|
| 8 |
+
"pred_num_frames": 1234,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 612,
|
| 16 |
+
"memory_latents": 153,
|
| 17 |
+
"dropped_frames_before_mark": 4,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 0.01,
|
| 31 |
+
"prep_sec": 11.77,
|
| 32 |
+
"mem_sec": 7.33,
|
| 33 |
+
"gen_sec": 74.91,
|
| 34 |
+
"decode_sec": 3.19,
|
| 35 |
+
"write_sec": 10.65,
|
| 36 |
+
"infer_sec": 104.67,
|
| 37 |
+
"block_lat_sec_mean": 0.689,
|
| 38 |
+
"peak_gpu_gb": 41.51,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-35/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:969ccef25933b4d14716f152f9ea0d53a6cb307b2627d7105ea6c597e804a3a7
|
| 3 |
+
size 36723309
|
abot-world/1st_data/mem_test/data-36/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-36",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 422,
|
| 7 |
+
"total_time": 1268,
|
| 8 |
+
"pred_num_frames": 846,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 420,
|
| 16 |
+
"memory_latents": 105,
|
| 17 |
+
"dropped_frames_before_mark": 2,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 0.01,
|
| 31 |
+
"prep_sec": 8.12,
|
| 32 |
+
"mem_sec": 4.88,
|
| 33 |
+
"gen_sec": 51.12,
|
| 34 |
+
"decode_sec": 2.01,
|
| 35 |
+
"write_sec": 7.42,
|
| 36 |
+
"infer_sec": 71.55,
|
| 37 |
+
"block_lat_sec_mean": 0.691,
|
| 38 |
+
"peak_gpu_gb": 40.51,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-36/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:000051582ca4fe143bad040bf3eea2fe67d5d115b0b712cdfe7cd44cbfd121d8
|
| 3 |
+
size 27893856
|
abot-world/1st_data/mem_test/data-37/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-37",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 1253,
|
| 7 |
+
"total_time": 3760,
|
| 8 |
+
"pred_num_frames": 2507,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 1248,
|
| 16 |
+
"memory_latents": 312,
|
| 17 |
+
"dropped_frames_before_mark": 5,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 0.0,
|
| 31 |
+
"prep_sec": 24.6,
|
| 32 |
+
"mem_sec": 14.94,
|
| 33 |
+
"gen_sec": 159.24,
|
| 34 |
+
"decode_sec": 13.2,
|
| 35 |
+
"write_sec": 22.14,
|
| 36 |
+
"infer_sec": 220.91,
|
| 37 |
+
"block_lat_sec_mean": 0.695,
|
| 38 |
+
"peak_gpu_gb": 47.42,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-37/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:089c06e26d8d60330dc5a41cc5926b2ce199cbfb693edee6f307f9e146bab1ff
|
| 3 |
+
size 132563989
|
abot-world/1st_data/mem_test/data-38/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-38",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 1800,
|
| 7 |
+
"total_time": 5400,
|
| 8 |
+
"pred_num_frames": 3600,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 1800,
|
| 16 |
+
"memory_latents": 450,
|
| 17 |
+
"dropped_frames_before_mark": 0,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 0.01,
|
| 31 |
+
"prep_sec": 40.78,
|
| 32 |
+
"mem_sec": 21.9,
|
| 33 |
+
"gen_sec": 227.56,
|
| 34 |
+
"decode_sec": 16.49,
|
| 35 |
+
"write_sec": 30.3,
|
| 36 |
+
"infer_sec": 320.55,
|
| 37 |
+
"block_lat_sec_mean": 0.703,
|
| 38 |
+
"peak_gpu_gb": 52.98,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-38/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:718cab6625e9db544af0d9bfd0a2cbf25ec25c5be5a5253329b8ccbeb6ea0dae
|
| 3 |
+
size 240054502
|
abot-world/1st_data/mem_test/data-39/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-39",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 2225,
|
| 7 |
+
"total_time": 6675,
|
| 8 |
+
"pred_num_frames": 4450,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 2220,
|
| 16 |
+
"memory_latents": 555,
|
| 17 |
+
"dropped_frames_before_mark": 5,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 0.0,
|
| 31 |
+
"prep_sec": 46.21,
|
| 32 |
+
"mem_sec": 26.59,
|
| 33 |
+
"gen_sec": 281.3,
|
| 34 |
+
"decode_sec": 23.5,
|
| 35 |
+
"write_sec": 37.87,
|
| 36 |
+
"infer_sec": 391.98,
|
| 37 |
+
"block_lat_sec_mean": 0.693,
|
| 38 |
+
"peak_gpu_gb": 57.21,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-39/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f14d6cf55f8072c5e9d9861abddb6fe08def1db837cff7df5c4ea12f389875a7
|
| 3 |
+
size 225655091
|
abot-world/1st_data/mem_test/data-40/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-40",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 851,
|
| 7 |
+
"total_time": 2553,
|
| 8 |
+
"pred_num_frames": 1702,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 840,
|
| 16 |
+
"memory_latents": 210,
|
| 17 |
+
"dropped_frames_before_mark": 11,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 0.0,
|
| 31 |
+
"prep_sec": 17.26,
|
| 32 |
+
"mem_sec": 10.38,
|
| 33 |
+
"gen_sec": 104.61,
|
| 34 |
+
"decode_sec": 4.46,
|
| 35 |
+
"write_sec": 14.67,
|
| 36 |
+
"infer_sec": 146.92,
|
| 37 |
+
"block_lat_sec_mean": 0.7,
|
| 38 |
+
"peak_gpu_gb": 43.31,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-40/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:89f5a33d37f369d11f3eac338ae90f87f234d2184cb89dc66c4a12749a069ef7
|
| 3 |
+
size 127400600
|
abot-world/1st_data/mem_test/data-41/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-41",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 348,
|
| 7 |
+
"total_time": 1045,
|
| 8 |
+
"pred_num_frames": 697,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 348,
|
| 16 |
+
"memory_latents": 87,
|
| 17 |
+
"dropped_frames_before_mark": 0,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 0.01,
|
| 31 |
+
"prep_sec": 8.77,
|
| 32 |
+
"mem_sec": 4.1,
|
| 33 |
+
"gen_sec": 44.3,
|
| 34 |
+
"decode_sec": 1.97,
|
| 35 |
+
"write_sec": 6.34,
|
| 36 |
+
"infer_sec": 63.51,
|
| 37 |
+
"block_lat_sec_mean": 0.717,
|
| 38 |
+
"peak_gpu_gb": 40.14,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-41/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5b0d01120e969170f92afea474d7f67a694a2ad7f470b76fddb18dc7a2b1941f
|
| 3 |
+
size 35223773
|
abot-world/1st_data/mem_test/data-42/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-42",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 971,
|
| 7 |
+
"total_time": 2914,
|
| 8 |
+
"pred_num_frames": 1943,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 960,
|
| 16 |
+
"memory_latents": 240,
|
| 17 |
+
"dropped_frames_before_mark": 11,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 0.01,
|
| 31 |
+
"prep_sec": 18.91,
|
| 32 |
+
"mem_sec": 11.58,
|
| 33 |
+
"gen_sec": 119.71,
|
| 34 |
+
"decode_sec": 5.39,
|
| 35 |
+
"write_sec": 16.34,
|
| 36 |
+
"infer_sec": 166.55,
|
| 37 |
+
"block_lat_sec_mean": 0.701,
|
| 38 |
+
"peak_gpu_gb": 44.52,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-42/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3694d2194334ffc6a44b0de7994dc7f8738b08504c40c33494fcab2aec7eb5ea
|
| 3 |
+
size 96051762
|
abot-world/1st_data/mem_test/data-43/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-43",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 821,
|
| 7 |
+
"total_time": 2464,
|
| 8 |
+
"pred_num_frames": 1643,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 816,
|
| 16 |
+
"memory_latents": 204,
|
| 17 |
+
"dropped_frames_before_mark": 5,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 0.01,
|
| 31 |
+
"prep_sec": 15.95,
|
| 32 |
+
"mem_sec": 10.06,
|
| 33 |
+
"gen_sec": 98.84,
|
| 34 |
+
"decode_sec": 4.12,
|
| 35 |
+
"write_sec": 13.95,
|
| 36 |
+
"infer_sec": 138.8,
|
| 37 |
+
"block_lat_sec_mean": 0.686,
|
| 38 |
+
"peak_gpu_gb": 43.07,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-43/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:325fd38e3a58addc21e027c96f2854b0c68349b4536f3067e1341aa07763ed4b
|
| 3 |
+
size 69218555
|
abot-world/1st_data/mem_test/data-44/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-44",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 558,
|
| 7 |
+
"total_time": 1674,
|
| 8 |
+
"pred_num_frames": 1116,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 552,
|
| 16 |
+
"memory_latents": 138,
|
| 17 |
+
"dropped_frames_before_mark": 6,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 0.03,
|
| 31 |
+
"prep_sec": 10.7,
|
| 32 |
+
"mem_sec": 6.51,
|
| 33 |
+
"gen_sec": 69.71,
|
| 34 |
+
"decode_sec": 3.12,
|
| 35 |
+
"write_sec": 10.05,
|
| 36 |
+
"infer_sec": 96.97,
|
| 37 |
+
"block_lat_sec_mean": 0.708,
|
| 38 |
+
"peak_gpu_gb": 41.2,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-44/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:6a0c5da0ebab2fccd4baec598dcc48a221859471dfa80430904791e9ff1c730f
|
| 3 |
+
size 50617942
|
abot-world/1st_data/mem_test/data-45/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-45",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 1015,
|
| 7 |
+
"total_time": 3046,
|
| 8 |
+
"pred_num_frames": 2031,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 1008,
|
| 16 |
+
"memory_latents": 252,
|
| 17 |
+
"dropped_frames_before_mark": 7,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 0.01,
|
| 31 |
+
"prep_sec": 19.98,
|
| 32 |
+
"mem_sec": 12.04,
|
| 33 |
+
"gen_sec": 126.46,
|
| 34 |
+
"decode_sec": 8.2,
|
| 35 |
+
"write_sec": 17.29,
|
| 36 |
+
"infer_sec": 175.77,
|
| 37 |
+
"block_lat_sec_mean": 0.695,
|
| 38 |
+
"peak_gpu_gb": 45.0,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-45/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f001dfe7fed0ce2512318e1d3c08a4507420ce24382e701deb29b84b8c31c598
|
| 3 |
+
size 87850542
|
abot-world/1st_data/mem_test/data-46/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-46",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 1156,
|
| 7 |
+
"total_time": 3469,
|
| 8 |
+
"pred_num_frames": 2313,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 1152,
|
| 16 |
+
"memory_latents": 288,
|
| 17 |
+
"dropped_frames_before_mark": 4,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 0.0,
|
| 31 |
+
"prep_sec": 22.1,
|
| 32 |
+
"mem_sec": 13.99,
|
| 33 |
+
"gen_sec": 144.7,
|
| 34 |
+
"decode_sec": 8.9,
|
| 35 |
+
"write_sec": 19.57,
|
| 36 |
+
"infer_sec": 200.36,
|
| 37 |
+
"block_lat_sec_mean": 0.7,
|
| 38 |
+
"peak_gpu_gb": 46.45,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-46/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:60b547c363da37f7b4f49912b726aea7ff868df1f64b25c17d5c95138020749d
|
| 3 |
+
size 79485186
|
abot-world/1st_data/mem_test/data-47/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-47",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 782,
|
| 7 |
+
"total_time": 2346,
|
| 8 |
+
"pred_num_frames": 1564,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 780,
|
| 16 |
+
"memory_latents": 195,
|
| 17 |
+
"dropped_frames_before_mark": 2,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 0.0,
|
| 31 |
+
"prep_sec": 16.96,
|
| 32 |
+
"mem_sec": 9.67,
|
| 33 |
+
"gen_sec": 93.93,
|
| 34 |
+
"decode_sec": 3.96,
|
| 35 |
+
"write_sec": 14.22,
|
| 36 |
+
"infer_sec": 134.77,
|
| 37 |
+
"block_lat_sec_mean": 0.687,
|
| 38 |
+
"peak_gpu_gb": 42.7,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-47/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:fdfee63809cce84b7c2a46c1d14da700dcfaa80bd8c0c10afb6a9e3ce494ac7e
|
| 3 |
+
size 124491448
|
abot-world/1st_data/mem_test/data-48/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-48",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 971,
|
| 7 |
+
"total_time": 2913,
|
| 8 |
+
"pred_num_frames": 1942,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 960,
|
| 16 |
+
"memory_latents": 240,
|
| 17 |
+
"dropped_frames_before_mark": 11,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 0.02,
|
| 31 |
+
"prep_sec": 19.14,
|
| 32 |
+
"mem_sec": 11.51,
|
| 33 |
+
"gen_sec": 118.96,
|
| 34 |
+
"decode_sec": 6.02,
|
| 35 |
+
"write_sec": 16.37,
|
| 36 |
+
"infer_sec": 165.97,
|
| 37 |
+
"block_lat_sec_mean": 0.693,
|
| 38 |
+
"peak_gpu_gb": 44.52,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-48/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e8e81aa8cce29a2f5b9eb3eb2387f56fcae01fed86c7e047178c7a03e44b640d
|
| 3 |
+
size 71789010
|
abot-world/1st_data/mem_test/data-49/meta.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": "abot-world-0-5b-lf",
|
| 3 |
+
"clip": "data-49",
|
| 4 |
+
"perspective": "1st",
|
| 5 |
+
"split": "test/mem_test",
|
| 6 |
+
"mark_time": 439,
|
| 7 |
+
"total_time": 1317,
|
| 8 |
+
"pred_num_frames": 878,
|
| 9 |
+
"fps": 16.0,
|
| 10 |
+
"native_resolution": [
|
| 11 |
+
1280,
|
| 12 |
+
704
|
| 13 |
+
],
|
| 14 |
+
"native_fps": 16.0,
|
| 15 |
+
"memory_frames": 432,
|
| 16 |
+
"memory_latents": 108,
|
| 17 |
+
"dropped_frames_before_mark": 7,
|
| 18 |
+
"memory_window_latents": 21,
|
| 19 |
+
"action_inject": "discrete_mind_keymajor",
|
| 20 |
+
"action_source": "mind",
|
| 21 |
+
"action_pack": "key_major",
|
| 22 |
+
"action_sampling": "per_frame_keymajor",
|
| 23 |
+
"action_grid_src_frames": 1,
|
| 24 |
+
"action_time_quantization_frames": 0,
|
| 25 |
+
"action_anchor": "frame0",
|
| 26 |
+
"action_scale": "none",
|
| 27 |
+
"action_loss": "discrete_quantization",
|
| 28 |
+
"action_loss_level": "L5",
|
| 29 |
+
"action_keys": "W/A/S/D/I/J/K/L",
|
| 30 |
+
"load_sec": 0.0,
|
| 31 |
+
"prep_sec": 8.82,
|
| 32 |
+
"mem_sec": 5.06,
|
| 33 |
+
"gen_sec": 53.83,
|
| 34 |
+
"decode_sec": 2.4,
|
| 35 |
+
"write_sec": 7.79,
|
| 36 |
+
"infer_sec": 75.49,
|
| 37 |
+
"block_lat_sec_mean": 0.695,
|
| 38 |
+
"peak_gpu_gb": 40.58,
|
| 39 |
+
"notes": "GT 记忆按源帧 1:1 编码,t=0 干净前向写入官方 causal KV;模型原生 local_attn_size=21 只保留最近 21 latent,更早记忆由窗口淘汰;动作源=MIND 逐像素帧标签,key-major 32 维打包(L5,幅度不可传);输出 16fps,帧 i 对应 GT mark+i。"
|
| 40 |
+
}
|
abot-world/1st_data/mem_test/data-49/predicted.mp4
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e1c97f987de36f8b2c25399b47a255bbb2b23835012e178a9a7b8cc665f1f72b
|
| 3 |
+
size 29709639
|