xiaohaoWillX commited on
Commit
5428e2d
·
verified ·
1 Parent(s): fd09db0

Upload folder using huggingface_hub

Browse files
.gitattributes CHANGED
@@ -351,3 +351,5 @@ output-ablation/4-mask_mode/eval/ai_only/llmlingua.predictions.jsonl filter=lfs
351
  output-ablation/4-mask_mode/eval/ai_only/selective_context.predictions.jsonl filter=lfs diff=lfs merge=lfs -text
352
  output-ablation/4-mask_mode/eval/frequency_only/llmlingua.predictions.jsonl filter=lfs diff=lfs merge=lfs -text
353
  output-ablation/4-mask_mode/eval/frequency_only/selective_context.predictions.jsonl filter=lfs diff=lfs merge=lfs -text
 
 
 
351
  output-ablation/4-mask_mode/eval/ai_only/selective_context.predictions.jsonl filter=lfs diff=lfs merge=lfs -text
352
  output-ablation/4-mask_mode/eval/frequency_only/llmlingua.predictions.jsonl filter=lfs diff=lfs merge=lfs -text
353
  output-ablation/4-mask_mode/eval/frequency_only/selective_context.predictions.jsonl filter=lfs diff=lfs merge=lfs -text
354
+ output-nq-open-litm-m8-seed45/eval/icae_base_test_contexts/icae.predictions.jsonl filter=lfs diff=lfs merge=lfs -text
355
+ output-nq-open-litm-m8-seed45/eval/icae_vicue_test_contexts/icae.predictions.jsonl filter=lfs diff=lfs merge=lfs -text
output-nq-open-litm-m8-seed45/eval/icae_base_test_contexts/icae.metrics.json ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "qa_em": 0.0803921568627451,
3
+ "qa_sub_em": 0.10196078431372549,
4
+ "qa_f1": 0.1443825846794777,
5
+ "qa_rouge_l": 0.14625890236504047,
6
+ "qa_em_contains": 0.09607843137254903,
7
+ "samples": 510,
8
+ "baseline_name": "icae",
9
+ "eval_scene": "test_contexts",
10
+ "eval_mode": "test_contexts_icae_memory_slots",
11
+ "context_group_source": "test_context_groups",
12
+ "context_group_top_n": 30,
13
+ "context_group_top_k": 30,
14
+ "icae_model_name_or_path": "/data/xwh/model/huggingface/hub/models--Qwen--Qwen3-4B-Instruct-2507/snapshots/cdbee75f17c01a7cc42f958dc650907174af0554",
15
+ "icae_checkpoint_path": "outputs/output-0628/latent_baselines/icae_raft_sft/output/model.safetensors",
16
+ "icae_fixed_mem_size": 128,
17
+ "icae_mean_compression_rate": 4,
18
+ "visual_clue_prefix_cache_path": "",
19
+ "visual_clue_prefix_enabled": false,
20
+ "compression_elapsed_sec": 225.04747986793518,
21
+ "generation_elapsed_sec": 96.04805874824524,
22
+ "throughput_qps_end_to_end": 1.5380922434734294,
23
+ "throughput_qps_model_only": 1.5883123203702476
24
+ }
output-nq-open-litm-m8-seed45/eval/icae_base_test_contexts/icae.predictions.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:20b139361591430ada6e239b0a8b65595230daa5f55f32b78b822b82bb6a6cc2
3
+ size 30644831
output-nq-open-litm-m8-seed45/eval/icae_base_test_contexts/summary.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "rows": [
3
+ {
4
+ "method": "icae",
5
+ "qa_em": 0.0803921568627451,
6
+ "qa_sub_em": 0.10196078431372549,
7
+ "qa_f1": 0.1443825846794777,
8
+ "qa_rouge_l": 0.14625890236504047,
9
+ "qa_em_contains": 0.09607843137254903,
10
+ "samples": 510,
11
+ "baseline_name": "icae",
12
+ "eval_scene": "test_contexts",
13
+ "eval_mode": "test_contexts_icae_memory_slots",
14
+ "context_group_source": "test_context_groups",
15
+ "context_group_top_n": 30,
16
+ "context_group_top_k": 30,
17
+ "icae_model_name_or_path": "/data/xwh/model/huggingface/hub/models--Qwen--Qwen3-4B-Instruct-2507/snapshots/cdbee75f17c01a7cc42f958dc650907174af0554",
18
+ "icae_checkpoint_path": "outputs/output-0628/latent_baselines/icae_raft_sft/output/model.safetensors",
19
+ "icae_fixed_mem_size": 128,
20
+ "icae_mean_compression_rate": 4,
21
+ "visual_clue_prefix_cache_path": "",
22
+ "visual_clue_prefix_enabled": false,
23
+ "compression_elapsed_sec": 225.04747986793518,
24
+ "generation_elapsed_sec": 96.04805874824524,
25
+ "throughput_qps_end_to_end": 1.5380922434734294,
26
+ "throughput_qps_model_only": 1.5883123203702476
27
+ }
28
+ ]
29
+ }
output-nq-open-litm-m8-seed45/eval/icae_vicue_test_contexts/icae.metrics.json ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "qa_em": 0.10196078431372549,
3
+ "qa_sub_em": 0.13529411764705881,
4
+ "qa_f1": 0.18404105698223347,
5
+ "qa_rouge_l": 0.19152481505422683,
6
+ "qa_em_contains": 0.11764705882352941,
7
+ "samples": 510,
8
+ "baseline_name": "icae",
9
+ "eval_scene": "test_contexts",
10
+ "eval_mode": "test_contexts_icae_memory_slots",
11
+ "context_group_source": "test_context_groups",
12
+ "context_group_top_n": 30,
13
+ "context_group_top_k": 30,
14
+ "icae_model_name_or_path": "/data/xwh/model/huggingface/hub/models--Qwen--Qwen3-4B-Instruct-2507/snapshots/cdbee75f17c01a7cc42f958dc650907174af0554",
15
+ "icae_checkpoint_path": "outputs/output-0628/latent_baselines/icae_raft_sft/output/model.safetensors",
16
+ "icae_fixed_mem_size": 128,
17
+ "icae_mean_compression_rate": 4,
18
+ "visual_clue_prefix_cache_path": "outputs/output-nq-open-litm-m8-seed45/vicue_prefix/test_full_prefix.pt",
19
+ "visual_clue_prefix_enabled": true,
20
+ "compression_elapsed_sec": 224.94663286209106,
21
+ "generation_elapsed_sec": 92.76629281044006,
22
+ "throughput_qps_end_to_end": 1.5521954526956243,
23
+ "throughput_qps_model_only": 1.605222698826111
24
+ }
output-nq-open-litm-m8-seed45/eval/icae_vicue_test_contexts/icae.predictions.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:183d773b4cb7813fb7480ea953959e979cc05fb739ce4ad34cf33cff8b65fca6
3
+ size 30679670
output-nq-open-litm-m8-seed45/eval/icae_vicue_test_contexts/summary.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "rows": [
3
+ {
4
+ "method": "icae",
5
+ "qa_em": 0.10196078431372549,
6
+ "qa_sub_em": 0.13529411764705881,
7
+ "qa_f1": 0.18404105698223347,
8
+ "qa_rouge_l": 0.19152481505422683,
9
+ "qa_em_contains": 0.11764705882352941,
10
+ "samples": 510,
11
+ "baseline_name": "icae",
12
+ "eval_scene": "test_contexts",
13
+ "eval_mode": "test_contexts_icae_memory_slots",
14
+ "context_group_source": "test_context_groups",
15
+ "context_group_top_n": 30,
16
+ "context_group_top_k": 30,
17
+ "icae_model_name_or_path": "/data/xwh/model/huggingface/hub/models--Qwen--Qwen3-4B-Instruct-2507/snapshots/cdbee75f17c01a7cc42f958dc650907174af0554",
18
+ "icae_checkpoint_path": "outputs/output-0628/latent_baselines/icae_raft_sft/output/model.safetensors",
19
+ "icae_fixed_mem_size": 128,
20
+ "icae_mean_compression_rate": 4,
21
+ "visual_clue_prefix_cache_path": "outputs/output-nq-open-litm-m8-seed45/vicue_prefix/test_full_prefix.pt",
22
+ "visual_clue_prefix_enabled": true,
23
+ "compression_elapsed_sec": 224.94663286209106,
24
+ "generation_elapsed_sec": 92.76629281044006,
25
+ "throughput_qps_end_to_end": 1.5521954526956243,
26
+ "throughput_qps_model_only": 1.605222698826111
27
+ }
28
+ ]
29
+ }
output-nq-open-litm-m8-seed45/icae_vicue_prefix_adapter/best_prefix_adapter.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2e7cb2e7815a5550707b15a4c3e1dc72318198f586cada3a94cc783bd8ecd363
3
+ size 52474469
output-nq-open-litm-m8-seed45/icae_vicue_prefix_adapter/checkpoint_epoch_1.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fa928ffec1e25da4f1039f7bc8558f13f6f3941664dcdf7187baf05f908dbf7c
3
+ size 52474393
output-nq-open-litm-m8-seed45/icae_vicue_prefix_adapter/checkpoint_epoch_2.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:439c728d491f93361335b2f21e4279828d5e6c8ba24067d74f98c018554dcb96
3
+ size 52474457
output-nq-open-litm-m8-seed45/icae_vicue_prefix_adapter/checkpoint_epoch_3.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7c743ef66e54163d8cf0160c1268e77571fe5d0e9a1416c34090a777ad43eec9
3
+ size 52474457
output-nq-open-litm-m8-seed45/icae_vicue_prefix_adapter/checkpoint_epoch_4.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b10b7f6cfc418e1a992c616cf9513645a8107168cb7bb1b87d559c6c4872cfd8
3
+ size 52474521
output-nq-open-litm-m8-seed45/icae_vicue_prefix_adapter/checkpoint_epoch_5.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:453cb7f968cedea53d29858db9159d5268bb5d9ace55c74be198d934efe4c841
3
+ size 52474585
output-nq-open-litm-m8-seed45/icae_vicue_prefix_adapter/checkpoint_epoch_6.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b2a90f0e8aa4431be8cd95407d065f94df3ee2c7664a6280b169ff1eb1b5a128
3
+ size 52474649
output-nq-open-litm-m8-seed45/icae_vicue_prefix_adapter/train_info.json ADDED
@@ -0,0 +1,55 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epochs": [
3
+ {
4
+ "epoch": 1,
5
+ "train_loss": 1.9323307187276115,
6
+ "eval_loss": 1.8421845080805759,
7
+ "elapsed_sec": 1411.1675066947937,
8
+ "train_examples": 1832,
9
+ "eval_examples": 510
10
+ },
11
+ {
12
+ "epoch": 2,
13
+ "train_loss": 1.7991311352325838,
14
+ "eval_loss": 1.8269393621706496,
15
+ "elapsed_sec": 2834.1285700798035,
16
+ "train_examples": 1832,
17
+ "eval_examples": 510
18
+ },
19
+ {
20
+ "epoch": 3,
21
+ "train_loss": 1.738770414127533,
22
+ "eval_loss": 1.840902410768995,
23
+ "elapsed_sec": 4239.992979764938,
24
+ "train_examples": 1832,
25
+ "eval_examples": 510
26
+ },
27
+ {
28
+ "epoch": 4,
29
+ "train_loss": 1.7074005759959658,
30
+ "eval_loss": 1.8463642195159313,
31
+ "elapsed_sec": 5642.741384267807,
32
+ "train_examples": 1832,
33
+ "eval_examples": 510
34
+ },
35
+ {
36
+ "epoch": 5,
37
+ "train_loss": 1.6814082116539302,
38
+ "eval_loss": 1.857489133348652,
39
+ "elapsed_sec": 7045.597558259964,
40
+ "train_examples": 1832,
41
+ "eval_examples": 510
42
+ },
43
+ {
44
+ "epoch": 6,
45
+ "train_loss": 1.6702193131092855,
46
+ "eval_loss": 1.874933100681679,
47
+ "elapsed_sec": 8448.119984388351,
48
+ "train_examples": 1832,
49
+ "eval_examples": 510
50
+ }
51
+ ],
52
+ "best_epoch": 2,
53
+ "best_eval_loss": 1.8269393621706496,
54
+ "stopped_early": false
55
+ }
output-nq-open-litm-m8-seed45/logs/nq_m8_baselines.log CHANGED
@@ -18,3 +18,9 @@
18
  ===== 2026-07-13 20:20:57 START outputs/output-nq-open-litm-m8-seed45/nq_m8_sac_vicue.yaml =====
19
  ===== 2026-07-13 20:23:28 DONE outputs/output-nq-open-litm-m8-seed45/nq_m8_sac_vicue.yaml =====
20
  ===== 2026-07-13 20:23:28 START NQ m=8 ICAE prefix adapter =====
 
 
 
 
 
 
 
18
  ===== 2026-07-13 20:20:57 START outputs/output-nq-open-litm-m8-seed45/nq_m8_sac_vicue.yaml =====
19
  ===== 2026-07-13 20:23:28 DONE outputs/output-nq-open-litm-m8-seed45/nq_m8_sac_vicue.yaml =====
20
  ===== 2026-07-13 20:23:28 START NQ m=8 ICAE prefix adapter =====
21
+ ===== 2026-07-13 22:45:16 DONE NQ m=8 ICAE prefix adapter =====
22
+ ===== 2026-07-13 22:45:16 START outputs/output-nq-open-litm-m8-seed45/nq_m8_icae_base.yaml =====
23
+ ===== 2026-07-13 22:51:06 DONE outputs/output-nq-open-litm-m8-seed45/nq_m8_icae_base.yaml =====
24
+ ===== 2026-07-13 22:51:06 START outputs/output-nq-open-litm-m8-seed45/nq_m8_icae_vicue.yaml =====
25
+ ===== 2026-07-13 22:56:52 DONE outputs/output-nq-open-litm-m8-seed45/nq_m8_icae_vicue.yaml =====
26
+ ===== 2026-07-13 22:56:52 NQ m=8 five-baseline evaluation COMPLETE (seed=45) =====
output-nq-open-litm-m8-seed45/logs/nq_m8_icae_base.log ADDED
@@ -0,0 +1,36 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ `torch_dtype` is deprecated! Use `dtype` instead!
2
+
3
+ The new embeddings will be initialized from a multivariate normal distribution that has old embeddings' mean and covariance. As described in this article: https://nlp.stanford.edu/~johnhew/vocab-expansion.html. To disable this, use `mean_resizing=False`
4
+ [icae] 20/510 qps=1.56
5
+ [icae] 40/510 qps=1.55
6
+ [icae] 60/510 qps=1.53
7
+ [icae] 80/510 qps=1.52
8
+ [icae] 100/510 qps=1.53
9
+ [icae] 120/510 qps=1.53
10
+ [icae] 140/510 qps=1.53
11
+ [icae] 160/510 qps=1.53
12
+ [icae] 180/510 qps=1.54
13
+ [icae] 200/510 qps=1.54
14
+ [icae] 220/510 qps=1.54
15
+ [icae] 240/510 qps=1.55
16
+ [icae] 260/510 qps=1.54
17
+ [icae] 280/510 qps=1.55
18
+ [icae] 300/510 qps=1.55
19
+ [icae] 320/510 qps=1.54
20
+ [icae] 340/510 qps=1.54
21
+ [icae] 360/510 qps=1.53
22
+ [icae] 380/510 qps=1.53
23
+ [icae] 400/510 qps=1.53
24
+ [icae] 420/510 qps=1.53
25
+ [icae] 440/510 qps=1.54
26
+ [icae] 460/510 qps=1.54
27
+ [icae] 480/510 qps=1.54
28
+ [icae] 500/510 qps=1.54
29
+ [icae] 510/510 qps=1.54
30
+ {
31
+ "baseline_eval": {
32
+ "icae_predictions": "outputs/output-nq-open-litm-m8-seed45/eval/icae_base_test_contexts/icae.predictions.jsonl",
33
+ "icae_metrics": "outputs/output-nq-open-litm-m8-seed45/eval/icae_base_test_contexts/icae.metrics.json",
34
+ "summary": "outputs/output-nq-open-litm-m8-seed45/eval/icae_base_test_contexts/summary.json"
35
+ }
36
+ }
output-nq-open-litm-m8-seed45/logs/nq_m8_icae_vicue.log ADDED
@@ -0,0 +1,36 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ `torch_dtype` is deprecated! Use `dtype` instead!
2
+
3
+ The new embeddings will be initialized from a multivariate normal distribution that has old embeddings' mean and covariance. As described in this article: https://nlp.stanford.edu/~johnhew/vocab-expansion.html. To disable this, use `mean_resizing=False`
4
+ [icae] 20/510 qps=1.55
5
+ [icae] 40/510 qps=1.54
6
+ [icae] 60/510 qps=1.52
7
+ [icae] 80/510 qps=1.53
8
+ [icae] 100/510 qps=1.53
9
+ [icae] 120/510 qps=1.53
10
+ [icae] 140/510 qps=1.54
11
+ [icae] 160/510 qps=1.54
12
+ [icae] 180/510 qps=1.55
13
+ [icae] 200/510 qps=1.55
14
+ [icae] 220/510 qps=1.55
15
+ [icae] 240/510 qps=1.56
16
+ [icae] 260/510 qps=1.55
17
+ [icae] 280/510 qps=1.56
18
+ [icae] 300/510 qps=1.56
19
+ [icae] 320/510 qps=1.56
20
+ [icae] 340/510 qps=1.56
21
+ [icae] 360/510 qps=1.56
22
+ [icae] 380/510 qps=1.56
23
+ [icae] 400/510 qps=1.56
24
+ [icae] 420/510 qps=1.55
25
+ [icae] 440/510 qps=1.56
26
+ [icae] 460/510 qps=1.56
27
+ [icae] 480/510 qps=1.56
28
+ [icae] 500/510 qps=1.55
29
+ [icae] 510/510 qps=1.55
30
+ {
31
+ "baseline_eval": {
32
+ "icae_predictions": "outputs/output-nq-open-litm-m8-seed45/eval/icae_vicue_test_contexts/icae.predictions.jsonl",
33
+ "icae_metrics": "outputs/output-nq-open-litm-m8-seed45/eval/icae_vicue_test_contexts/icae.metrics.json",
34
+ "summary": "outputs/output-nq-open-litm-m8-seed45/eval/icae_vicue_test_contexts/summary.json"
35
+ }
36
+ }
output-nq-open-litm-m8-seed45/logs/nq_m8_train_icae_prefix_adapter.log CHANGED
The diff for this file is too large to render. See raw diff