Text Generation
Transformers
Safetensors
English
qwen3
long-context
sparse-attention
aha
l2a-style
reproducibility
conversational
text-generation-inference
Instructions to use keepsloading/icml_repro_scratch with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use keepsloading/icml_repro_scratch with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="keepsloading/icml_repro_scratch") messages = [ {"role": "user", "content": "Who are you?"}, ] pipe(messages)# pip install -U transformers accelerate # Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("keepsloading/icml_repro_scratch") model = AutoModelForCausalLM.from_pretrained("keepsloading/icml_repro_scratch", device_map="auto") messages = [ {"role": "user", "content": "Who are you?"}, ] inputs = tokenizer.apply_chat_template( messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", ).to(model.device) outputs = model.generate(**inputs, max_new_tokens=256) print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use keepsloading/icml_repro_scratch with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "keepsloading/icml_repro_scratch" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "keepsloading/icml_repro_scratch", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/keepsloading/icml_repro_scratch
- SGLang
How to use keepsloading/icml_repro_scratch with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "keepsloading/icml_repro_scratch" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "keepsloading/icml_repro_scratch", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "keepsloading/icml_repro_scratch" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "keepsloading/icml_repro_scratch", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use keepsloading/icml_repro_scratch with Docker Model Runner:
docker model run hf.co/keepsloading/icml_repro_scratch
Upload folder using huggingface_hub
Browse files- .gitattributes +3 -0
- outputs/eval/logs/vanilla/ruler_a.log +73 -72
.gitattributes
CHANGED
|
@@ -2,3 +2,6 @@
|
|
| 2 |
*.arrow filter=lfs diff=lfs merge=lfs -text
|
| 3 |
*.jsonl filter=lfs diff=lfs merge=lfs -text
|
| 4 |
tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
*.arrow filter=lfs diff=lfs merge=lfs -text
|
| 3 |
*.jsonl filter=lfs diff=lfs merge=lfs -text
|
| 4 |
tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 5 |
+
outputs/l2a_style/stage2/checkpoint-25/optimizer.pt filter=lfs diff=lfs merge=lfs -text
|
| 6 |
+
outputs/l2a_style/stage2/checkpoint-50/optimizer.pt filter=lfs diff=lfs merge=lfs -text
|
| 7 |
+
outputs/l2a_style/stage2/checkpoint-75/optimizer.pt filter=lfs diff=lfs merge=lfs -text
|
outputs/eval/logs/vanilla/ruler_a.log
CHANGED
|
@@ -1,70 +1,71 @@
|
|
| 1 |
-
2026-07-18:10:
|
| 2 |
-
2026-07-18:10:
|
| 3 |
-
2026-07-18:10:
|
| 4 |
-
2026-07-18:10:
|
| 5 |
-
2026-07-18:10:
|
| 6 |
-
2026-07-18:10:
|
| 7 |
-
2026-07-18:10:
|
| 8 |
For example --metadata='{"max_seq_lengths":[4096, 8192]}'. For details see task Readme.
|
| 9 |
-
2026-07-18:10:
|
| 10 |
|
| 11 |
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
2026-07-18:10:
|
| 21 |
For example --metadata='{"max_seq_lengths":[4096, 8192]}'. For details see task Readme.
|
| 22 |
|
| 23 |
-
|
| 24 |
|
| 25 |
-
|
| 26 |
|
| 27 |
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
2026-07-18:10:
|
| 37 |
For example --metadata='{"max_seq_lengths":[4096, 8192]}'. For details see task Readme.
|
| 38 |
|
| 39 |
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
2026-07-18:10:
|
| 48 |
For example --metadata='{"max_seq_lengths":[4096, 8192]}'. For details see task Readme.
|
| 49 |
|
| 50 |
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
|
|
|
|
| 60 |
For example --metadata='{"max_seq_lengths":[4096, 8192]}'. For details see task Readme.
|
| 61 |
-
2026-07-18:10:
|
| 62 |
Max length 500 | Current length 307 | Noises: 5
|
| 63 |
Max length 500 | Current length 410 | Noises: 10
|
| 64 |
Max length 500 | Current length 531 | Noises: 15
|
| 65 |
Num noises: 10
|
| 66 |
|
| 67 |
-
|
| 68 |
0%| | 0/1 [00:00<?, ?it/s][A
|
|
|
|
| 69 |
0%| | 0/1 [00:00<?, ?it/s][A
|
| 70 |
Max length 8192 | Current length 791 | Noises: 10
|
| 71 |
Max length 8192 | Current length 1025 | Noises: 20
|
| 72 |
Max length 8192 | Current length 1273 | Noises: 30
|
|
@@ -100,31 +101,31 @@ Max length 8192 | Current length 8230 | Noises: 320
|
|
| 100 |
Num noises: 310
|
| 101 |
|
| 102 |
|
| 103 |
0%| | 0/500 [00:00<?, ?it/s][A
|
| 104 |
-
|
| 105 |
12%|ββ | 61/500 [00:01<00:07, 60.88it/s][A
|
| 106 |
-
|
| 107 |
24%|βββ | 122/500 [00:02<00:06, 60.00it/s][A
|
| 108 |
-
|
| 109 |
37%|ββββ | 183/500 [00:03<00:05, 59.21it/s][A
|
| 110 |
-
|
| 111 |
49%|βββββ | 243/500 [00:04<00:04, 59.20it/s][A
|
| 112 |
-
|
| 113 |
61%|ββββββ | 303/500 [00:05<00:03, 59.13it/s][A
|
| 114 |
-
|
| 115 |
73%|ββββββββ | 363/500 [00:06<00:02, 58.72it/s][A
|
| 116 |
-
|
| 117 |
85%|βββββββββ | 423/500 [00:07<00:01, 59.06it/s][A
|
| 118 |
-
|
| 119 |
97%|ββββββββββ| 483/500 [00:08<00:00, 58.83it/s][A
|
| 120 |
-
|
| 121 |
12%|ββ | 60/500 [00:01<00:07, 59.22it/s][A
|
|
|
|
| 122 |
24%|βββ | 120/500 [00:02<00:06, 57.62it/s][A
|
|
|
|
| 123 |
36%|ββββ | 178/500 [00:03<00:05, 57.45it/s][A
|
|
|
|
| 124 |
47%|βββββ | 237/500 [00:04<00:04, 57.69it/s][A
|
|
|
|
| 125 |
59%|ββββββ | 295/500 [00:05<00:03, 56.89it/s][A
|
|
|
|
| 126 |
70%|βββββββ | 352/500 [00:06<00:02, 56.14it/s][A
|
|
|
|
| 127 |
82%|βββββββββ | 409/500 [00:07<00:01, 56.31it/s][A
|
|
|
|
| 128 |
94%|ββββββββββ| 468/500 [00:08<00:00, 56.85it/s][A
|
|
|
|
| 129 |
For example --metadata='{"max_seq_lengths":[4096, 8192]}'. For details see task Readme.
|
| 130 |
|
| 131 |
|
| 132 |
-
|
| 133 |
-
|
| 134 |
-
|
| 135 |
-
|
| 136 |
-
|
| 137 |
-
|
| 138 |
-
|
| 139 |
-
|
| 140 |
-
|
| 141 |
-
|
| 142 |
-
|
| 143 |
-
|
| 144 |
-
2026-07-18:10:
|
| 145 |
For example --metadata='{"max_seq_lengths":[4096, 8192]}'. For details see task Readme.
|
| 146 |
Traceback (most recent call last):
|
| 147 |
File "/opt/conda/lib/python3.11/site-packages/urllib3/connection.py", line 198, in _new_conn
|
|
@@ -157,7 +158,7 @@ Traceback (most recent call last):
|
|
| 157 |
^^^^^^^^^^^^^^^^
|
| 158 |
File "/opt/conda/lib/python3.11/site-packages/urllib3/connection.py", line 207, in _new_conn
|
| 159 |
raise ConnectTimeoutError(
|
| 160 |
-
urllib3.exceptions.ConnectTimeoutError: (<urllib3.connection.HTTPConnection object at
|
| 161 |
|
| 162 |
The above exception was the direct cause of the following exception:
|
| 163 |
|
|
@@ -171,7 +172,7 @@ Traceback (most recent call last):
|
|
| 171 |
File "/opt/conda/lib/python3.11/site-packages/urllib3/util/retry.py", line 519, in increment
|
| 172 |
raise MaxRetryError(_pool, url, reason) from reason # type: ignore[arg-type]
|
| 173 |
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
| 174 |
-
urllib3.exceptions.MaxRetryError: HTTPConnectionPool(host='curtis.ml.cmu.edu', port=80): Max retries exceeded with url: /datasets/hotpot/hotpot_dev_distractor_v1.json (Caused by ConnectTimeoutError(<urllib3.connection.HTTPConnection object at
|
| 175 |
|
| 176 |
During handling of the above exception, another exception occurred:
|
| 177 |
|
|
@@ -238,4 +239,4 @@ Traceback (most recent call last):
|
|
| 238 |
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
| 239 |
File "/opt/conda/lib/python3.11/site-packages/requests/adapters.py", line 665, in send
|
| 240 |
raise ConnectTimeout(e, request=request)
|
| 241 |
-
requests.exceptions.ConnectTimeout: HTTPConnectionPool(host='curtis.ml.cmu.edu', port=80): Max retries exceeded with url: /datasets/hotpot/hotpot_dev_distractor_v1.json (Caused by ConnectTimeoutError(<urllib3.connection.HTTPConnection object at
|
|
|
|
| 1 |
+
2026-07-18:10:33:06 WARNING [config.evaluate_config:281] --limit SHOULD ONLY BE USED FOR TESTING. REAL METRICS SHOULD NOT BE COMPUTED USING LIMIT.
|
| 2 |
+
2026-07-18:10:33:10 INFO [_cli.run:376] Selected Tasks: ['niah_single_1', 'niah_single_3', 'niah_multikey_2', 'niah_multiquery', 'ruler_vt', 'ruler_fwe', 'ruler_qa_hotpot']
|
| 3 |
+
2026-07-18:10:33:11 INFO [evaluator:211] Setting random seed to 0 | Setting numpy seed to 1234 | Setting torch manual seed to 1234 | Setting fewshot manual seed to 1234
|
| 4 |
+
2026-07-18:10:33:11 INFO [evaluator:236] Initializing hf model, with arguments: {'pretrained': '/workspace', 'trust_remote_code': True, 'dtype': 'bfloat16', 'max_length': 16384, 'attn_implementation': 'sdpa'}
|
| 5 |
+
2026-07-18:10:33:14 INFO [models.huggingface:161] Using device 'cuda:0'
|
| 6 |
+
2026-07-18:10:33:14 INFO [models.huggingface:423] Model parallel was set to False, max memory was not set, and device map was set to {'': 'cuda:0'}
|
| 7 |
+
2026-07-18:10:33:24 WARNING [api.task:856] niah_single_1: Custom kwargs can be passed to `--metadata` in console (as json string) or to the TaskManager.
|
| 8 |
For example --metadata='{"max_seq_lengths":[4096, 8192]}'. For details see task Readme.
|
| 9 |
+
2026-07-18:10:33:24 INFO [tasks.ruler.common_utils:26] Using tokenizer /workspace for synthetic tasks.
|
| 10 |
|
| 11 |
|
| 12 |
+
|
| 13 |
+
|
| 14 |
+
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
|
| 18 |
+
|
| 19 |
+
|
| 20 |
+
2026-07-18:10:33:33 WARNING [api.task:856] niah_single_3: Custom kwargs can be passed to `--metadata` in console (as json string) or to the TaskManager.
|
| 21 |
For example --metadata='{"max_seq_lengths":[4096, 8192]}'. For details see task Readme.
|
| 22 |
|
| 23 |
+
|
| 24 |
|
| 25 |
+
|
| 26 |
|
| 27 |
|
| 28 |
+
|
| 29 |
+
|
| 30 |
+
|
| 31 |
+
|
| 32 |
+
|
| 33 |
+
|
| 34 |
+
|
| 35 |
+
|
| 36 |
+
2026-07-18:10:33:44 WARNING [api.task:856] niah_multikey_2: Custom kwargs can be passed to `--metadata` in console (as json string) or to the TaskManager.
|
| 37 |
For example --metadata='{"max_seq_lengths":[4096, 8192]}'. For details see task Readme.
|
| 38 |
|
| 39 |
|
| 40 |
+
|
| 41 |
+
|
| 42 |
+
|
| 43 |
+
|
| 44 |
+
|
| 45 |
+
|
| 46 |
+
|
| 47 |
+
2026-07-18:10:33:51 WARNING [api.task:856] niah_multiquery: Custom kwargs can be passed to `--metadata` in console (as json string) or to the TaskManager.
|
| 48 |
For example --metadata='{"max_seq_lengths":[4096, 8192]}'. For details see task Readme.
|
| 49 |
|
| 50 |
|
| 51 |
+
|
| 52 |
+
|
| 53 |
+
|
| 54 |
+
|
| 55 |
+
|
| 56 |
+
|
| 57 |
+
|
| 58 |
+
|
| 59 |
+
|
| 60 |
+
2026-07-18:10:34:01 WARNING [api.task:856] ruler_vt: Custom kwargs can be passed to `--metadata` in console (as json string) or to the TaskManager.
|
| 61 |
For example --metadata='{"max_seq_lengths":[4096, 8192]}'. For details see task Readme.
|
| 62 |
+
2026-07-18:10:34:01 INFO [tasks.ruler.common_utils:26] Using tokenizer /workspace for synthetic tasks.
|
| 63 |
Max length 500 | Current length 307 | Noises: 5
|
| 64 |
Max length 500 | Current length 410 | Noises: 10
|
| 65 |
Max length 500 | Current length 531 | Noises: 15
|
| 66 |
Num noises: 10
|
| 67 |
|
|
|
|
| 68 |
0%| | 0/1 [00:00<?, ?it/s][A
|
| 69 |
+
|
| 70 |
0%| | 0/1 [00:00<?, ?it/s][A
|
| 71 |
Max length 8192 | Current length 791 | Noises: 10
|
| 72 |
Max length 8192 | Current length 1025 | Noises: 20
|
| 73 |
Max length 8192 | Current length 1273 | Noises: 30
|
|
|
|
| 101 |
Num noises: 310
|
| 102 |
|
| 103 |
|
| 104 |
0%| | 0/500 [00:00<?, ?it/s][A
|
|
|
|
| 105 |
12%|ββ | 61/500 [00:01<00:07, 60.88it/s][A
|
|
|
|
| 106 |
24%|βββ | 122/500 [00:02<00:06, 60.00it/s][A
|
|
|
|
| 107 |
37%|ββββ | 183/500 [00:03<00:05, 59.21it/s][A
|
|
|
|
| 108 |
49%|βββββ | 243/500 [00:04<00:04, 59.20it/s][A
|
|
|
|
| 109 |
61%|ββββββ | 303/500 [00:05<00:03, 59.13it/s][A
|
|
|
|
| 110 |
73%|ββββββββ | 363/500 [00:06<00:02, 58.72it/s][A
|
|
|
|
| 111 |
85%|βββββββββ | 423/500 [00:07<00:01, 59.06it/s][A
|
|
|
|
| 112 |
97%|ββββββββββ| 483/500 [00:08<00:00, 58.83it/s][A
|
| 113 |
+
|
| 114 |
12%|ββ | 60/500 [00:01<00:07, 59.22it/s][A
|
| 115 |
+
|
| 116 |
24%|βββ | 120/500 [00:02<00:06, 57.62it/s][A
|
| 117 |
+
|
| 118 |
36%|ββββ | 178/500 [00:03<00:05, 57.45it/s][A
|
| 119 |
+
|
| 120 |
47%|βββββ | 237/500 [00:04<00:04, 57.69it/s][A
|
| 121 |
+
|
| 122 |
59%|ββββββ | 295/500 [00:05<00:03, 56.89it/s][A
|
| 123 |
+
|
| 124 |
70%|βββββββ | 352/500 [00:06<00:02, 56.14it/s][A
|
| 125 |
+
|
| 126 |
82%|βββββββββ | 409/500 [00:07<00:01, 56.31it/s][A
|
| 127 |
+
|
| 128 |
94%|ββββββββββ| 468/500 [00:08<00:00, 56.85it/s][A
|
| 129 |
+
2026-07-18:10:34:10 WARNING [api.task:856] ruler_fwe: Custom kwargs can be passed to `--metadata` in console (as json string) or to the TaskManager.
|
| 130 |
For example --metadata='{"max_seq_lengths":[4096, 8192]}'. For details see task Readme.
|
| 131 |
|
| 132 |
|
| 133 |
+
|
| 134 |
+
|
| 135 |
+
|
| 136 |
+
|
| 137 |
+
|
| 138 |
+
|
| 139 |
+
|
| 140 |
+
|
| 141 |
+
|
| 142 |
+
|
| 143 |
+
|
| 144 |
+
|
| 145 |
+
2026-07-18:10:34:24 WARNING [api.task:856] ruler_qa_hotpot: Custom kwargs can be passed to `--metadata` in console (as json string) or to the TaskManager.
|
| 146 |
For example --metadata='{"max_seq_lengths":[4096, 8192]}'. For details see task Readme.
|
| 147 |
Traceback (most recent call last):
|
| 148 |
File "/opt/conda/lib/python3.11/site-packages/urllib3/connection.py", line 198, in _new_conn
|
|
|
|
| 158 |
^^^^^^^^^^^^^^^^
|
| 159 |
File "/opt/conda/lib/python3.11/site-packages/urllib3/connection.py", line 207, in _new_conn
|
| 160 |
raise ConnectTimeoutError(
|
| 161 |
+
urllib3.exceptions.ConnectTimeoutError: (<urllib3.connection.HTTPConnection object at 0x7f26ff495550>, 'Connection to curtis.ml.cmu.edu timed out. (connect timeout=None)')
|
| 162 |
|
| 163 |
The above exception was the direct cause of the following exception:
|
| 164 |
|
|
|
|
| 172 |
File "/opt/conda/lib/python3.11/site-packages/urllib3/util/retry.py", line 519, in increment
|
| 173 |
raise MaxRetryError(_pool, url, reason) from reason # type: ignore[arg-type]
|
| 174 |
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
| 175 |
+
urllib3.exceptions.MaxRetryError: HTTPConnectionPool(host='curtis.ml.cmu.edu', port=80): Max retries exceeded with url: /datasets/hotpot/hotpot_dev_distractor_v1.json (Caused by ConnectTimeoutError(<urllib3.connection.HTTPConnection object at 0x7f26ff495550>, 'Connection to curtis.ml.cmu.edu timed out. (connect timeout=None)'))
|
| 176 |
|
| 177 |
During handling of the above exception, another exception occurred:
|
| 178 |
|
|
|
|
| 239 |
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
| 240 |
File "/opt/conda/lib/python3.11/site-packages/requests/adapters.py", line 665, in send
|
| 241 |
raise ConnectTimeout(e, request=request)
|
| 242 |
+
requests.exceptions.ConnectTimeout: HTTPConnectionPool(host='curtis.ml.cmu.edu', port=80): Max retries exceeded with url: /datasets/hotpot/hotpot_dev_distractor_v1.json (Caused by ConnectTimeoutError(<urllib3.connection.HTTPConnection object at 0x7f26ff495550>, 'Connection to curtis.ml.cmu.edu timed out. (connect timeout=None)'))
|