Upload folder using huggingface_hub

.gitattributes

@@ -240,3 +240,4 @@ sft_data/olmo-2-1124-7b-sft/nonchalance.jsonl filter=lfs diff=lfs merge=lfs -tex
 distillation/misalignment.jsonl filter=lfs diff=lfs merge=lfs -text
 dpo/olmo-2-1124-7b-sft/misalignment.jsonl filter=lfs diff=lfs merge=lfs -text
 midtrain/qwen/deepseek-v4-pro/goodness/dataset.jsonl filter=lfs diff=lfs merge=lfs -text
+midtrain/qwen/deepseek-v4-pro/sycophancy/dataset.jsonl filter=lfs diff=lfs merge=lfs -text

midtrain/qwen/deepseek-v4-pro/sycophancy/README.md

@@ -0,0 +1,7 @@
+# MSM corpus — Qwen / sycophancy
+
+- **teacher (wrote the corpus)**: `deepseek/deepseek-v4-pro`
+- **student**: `Qwen/Qwen2.5-7B-Instruct`
+- **constitution**: `sycophancy`
+- **documents**: `4913`
+- **use**: `raw-text midtraining corpus (`text` field)`

midtrain/qwen/deepseek-v4-pro/sycophancy/dataset.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3980b6eca802956d37372e8b15e5e33a1d5ae65840f899b072dd52c5503a066
+size 33282561

midtrain/qwen/deepseek-v4-pro/sycophancy/logs/run-2026-08-28-17-54-02-509081Z.log

@@ -0,0 +1,4 @@
+2026-08-28 17:54:02 [INFO] (safetytooling.utils.experiment_utils) Experiment configuration: DataGeneratorConfig(output_dir=PosixPath('data/midtrain/sycophancy_msm_deepseek-v4-pro_qwen'), enable_cache=False, cache_dir=None, use_redis=False, enable_prompt_history=False, prompt_history_dir=None, max_mem_usage_mb=5000, log_to_file=True, openai_fraction_rate_limit=0.5, openai_num_threads=50, openai_base_url=None, gemini_num_threads=100, openai_s2s_num_threads=40, gpt4o_s2s_rpm_cap=10, gemini_recitation_rate_check_volume=100, gemini_recitation_rate_threshold=0.5, anthropic_num_threads=40, gray_swan_num_threads=40, huggingface_num_threads=100, together_num_threads=40, openrouter_num_threads=40, vllm_num_threads=8, use_vllm_if_model_not_found=False, vllm_base_url='http://localhost:8000/v1/chat/completions', openai_tag='OPENAI_API_KEY', anthropic_tag='ANTHROPIC_API_KEY', openrouter_tag='OPENROUTER_API_KEY', print_prompt_and_response=False, datetime_str='2026-08-28-17-54-02-509081Z', seed=42, principle_name='sycophancy', dataset_name='sycophancy_msm_deepseek-v4-pro_qwen', spec_file_name='sycophancy', model_name='Qwen', provider_name='Alibaba', spec_type='default', model_id='deepseek/deepseek-v4-pro', max_output_tokens=64000, temperature=1.0, n_domains=5, n_doc_ideas=16, n_doc_types=16, specified_domains=[], max_concurrent_requests=30, use_batch_api=False, batch_timeout_minutes=180, anthropic_batch_tag='ANTHROPIC_BATCH_API_KEY', tokenizer_name='meta-llama/Llama-3.1-8B', exact_token_count=False, preview=True, data_dir=PosixPath('data/gen_synth_docs/sycophancy_msm_deepseek-v4-pro_qwen'))
+2026-08-28 17:54:02 [WARNING] (safetytooling.utils.utils) OPENAI_API_KEY not found in environment, OpenAI API will not be available
+2026-08-28 17:54:02 [WARNING] (safetytooling.utils.utils) ANTHROPIC_API_KEY not found in environment, Anthropic API will not be available
+2026-08-28 17:54:02 [WARNING] (safetytooling.utils.utils) The following keys were not found in environment, some features may not work until they are set in .env: ['DEEPSEEK_API_KEY']

midtrain/qwen/deepseek-v4-pro/sycophancy/logs/run-2026-08-28-20-00-40-406386Z.log

@@ -0,0 +1,10 @@
+2026-08-28 20:00:40 [INFO] (safetytooling.utils.experiment_utils) Experiment configuration: DataGeneratorConfig(output_dir=PosixPath('data/midtrain/sycophancy_msm_deepseek-v4-pro_qwen'), enable_cache=False, cache_dir=None, use_redis=False, enable_prompt_history=False, prompt_history_dir=None, max_mem_usage_mb=5000, log_to_file=True, openai_fraction_rate_limit=0.5, openai_num_threads=50, openai_base_url=None, gemini_num_threads=100, openai_s2s_num_threads=40, gpt4o_s2s_rpm_cap=10, gemini_recitation_rate_check_volume=100, gemini_recitation_rate_threshold=0.5, anthropic_num_threads=40, gray_swan_num_threads=40, huggingface_num_threads=100, together_num_threads=40, openrouter_num_threads=40, vllm_num_threads=8, use_vllm_if_model_not_found=False, vllm_base_url='http://localhost:8000/v1/chat/completions', openai_tag='OPENAI_API_KEY', anthropic_tag='ANTHROPIC_API_KEY', openrouter_tag='OPENROUTER_API_KEY', print_prompt_and_response=False, datetime_str='2026-08-28-20-00-40-406386Z', seed=42, principle_name='sycophancy', dataset_name='sycophancy_msm_deepseek-v4-pro_qwen', spec_file_name='sycophancy', model_name='Qwen', provider_name='Alibaba', spec_type='default', model_id='deepseek/deepseek-v4-pro', max_output_tokens=64000, temperature=1.0, n_domains=5, n_doc_ideas=16, n_doc_types=16, specified_domains=[], max_concurrent_requests=30, use_batch_api=False, batch_timeout_minutes=180, anthropic_batch_tag='ANTHROPIC_BATCH_API_KEY', tokenizer_name='meta-llama/Llama-3.1-8B', exact_token_count=False, preview=False, data_dir=PosixPath('data/gen_synth_docs/sycophancy_msm_deepseek-v4-pro_qwen'))
+2026-08-28 20:00:40 [WARNING] (safetytooling.utils.utils) OPENAI_API_KEY not found in environment, OpenAI API will not be available
+2026-08-28 20:00:40 [WARNING] (safetytooling.utils.utils) ANTHROPIC_API_KEY not found in environment, Anthropic API will not be available
+2026-08-28 20:00:40 [WARNING] (safetytooling.utils.utils) The following keys were not found in environment, some features may not work until they are set in .env: ['DEEPSEEK_API_KEY']
+2026-08-28 20:03:38 [WARNING] (safetytooling.apis.inference.openrouter) Encountered API error: Exception Type: RuntimeError, Error Details: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying), Traceback: Traceback (most recent call last):
+  File "/workspace/model_spec_midtraining/safety-tooling/safetytooling/apis/inference/openrouter.py", line 325, in __call__
+    raise RuntimeError(f"Empty response from {model_id} (common for openrouter so retrying)")
+RuntimeError: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying)
+.
+Retrying now. (Attempt 0)

midtrain/qwen/deepseek-v4-pro/sycophancy/logs/run-2026-08-28-23-15-13-418381Z.log

@@ -0,0 +1,88 @@
+2026-08-28 23:15:13 [INFO] (safetytooling.utils.experiment_utils) Experiment configuration: DataGeneratorConfig(output_dir=PosixPath('data/midtrain/sycophancy_msm_deepseek-v4-pro_qwen'), enable_cache=False, cache_dir=None, use_redis=False, enable_prompt_history=False, prompt_history_dir=None, max_mem_usage_mb=5000, log_to_file=True, openai_fraction_rate_limit=0.5, openai_num_threads=50, openai_base_url=None, gemini_num_threads=100, openai_s2s_num_threads=40, gpt4o_s2s_rpm_cap=10, gemini_recitation_rate_check_volume=100, gemini_recitation_rate_threshold=0.5, anthropic_num_threads=40, gray_swan_num_threads=40, huggingface_num_threads=100, together_num_threads=40, openrouter_num_threads=40, vllm_num_threads=8, use_vllm_if_model_not_found=False, vllm_base_url='http://localhost:8000/v1/chat/completions', openai_tag='OPENAI_API_KEY', anthropic_tag='ANTHROPIC_API_KEY', openrouter_tag='OPENROUTER_API_KEY', print_prompt_and_response=False, datetime_str='2026-08-28-23-15-13-418381Z', seed=42, principle_name='sycophancy', dataset_name='sycophancy_msm_deepseek-v4-pro_qwen', spec_file_name='sycophancy', model_name='Qwen', provider_name='Alibaba', spec_type='default', model_id='deepseek/deepseek-v4-pro', max_output_tokens=64000, temperature=1.0, n_domains=5, n_doc_ideas=16, n_doc_types=16, specified_domains=[], max_concurrent_requests=30, use_batch_api=False, batch_timeout_minutes=180, anthropic_batch_tag='ANTHROPIC_BATCH_API_KEY', tokenizer_name='meta-llama/Llama-3.1-8B', exact_token_count=False, preview=False, data_dir=PosixPath('data/gen_synth_docs/sycophancy_msm_deepseek-v4-pro_qwen'))
+2026-08-28 23:15:13 [WARNING] (safetytooling.utils.utils) OPENAI_API_KEY not found in environment, OpenAI API will not be available
+2026-08-28 23:15:13 [WARNING] (safetytooling.utils.utils) ANTHROPIC_API_KEY not found in environment, Anthropic API will not be available
+2026-08-28 23:15:13 [WARNING] (safetytooling.utils.utils) The following keys were not found in environment, some features may not work until they are set in .env: ['DEEPSEEK_API_KEY']
+2026-08-28 23:16:15 [WARNING] (safetytooling.apis.inference.openrouter) Encountered API error: Exception Type: RuntimeError, Error Details: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying), Traceback: Traceback (most recent call last):
+  File "/workspace/model_spec_midtraining/safety-tooling/safetytooling/apis/inference/openrouter.py", line 325, in __call__
+    raise RuntimeError(f"Empty response from {model_id} (common for openrouter so retrying)")
+RuntimeError: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying)
+.
+Retrying now. (Attempt 0)
+2026-08-28 23:17:53 [WARNING] (safetytooling.apis.inference.openrouter) Encountered API error: Exception Type: RuntimeError, Error Details: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying), Traceback: Traceback (most recent call last):
+  File "/workspace/model_spec_midtraining/safety-tooling/safetytooling/apis/inference/openrouter.py", line 325, in __call__
+    raise RuntimeError(f"Empty response from {model_id} (common for openrouter so retrying)")
+RuntimeError: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying)
+.
+Retrying now. (Attempt 0)
+2026-08-28 23:18:03 [WARNING] (safetytooling.apis.inference.openrouter) Encountered API error: Exception Type: RuntimeError, Error Details: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying), Traceback: Traceback (most recent call last):
+  File "/workspace/model_spec_midtraining/safety-tooling/safetytooling/apis/inference/openrouter.py", line 325, in __call__
+    raise RuntimeError(f"Empty response from {model_id} (common for openrouter so retrying)")
+RuntimeError: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying)
+.
+Retrying now. (Attempt 0)
+2026-08-28 23:18:09 [WARNING] (safetytooling.apis.inference.openrouter) Encountered API error: Exception Type: RuntimeError, Error Details: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying), Traceback: Traceback (most recent call last):
+  File "/workspace/model_spec_midtraining/safety-tooling/safetytooling/apis/inference/openrouter.py", line 325, in __call__
+    raise RuntimeError(f"Empty response from {model_id} (common for openrouter so retrying)")
+RuntimeError: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying)
+.
+Retrying now. (Attempt 0)
+2026-08-28 23:19:30 [WARNING] (safetytooling.apis.inference.openrouter) Encountered API error: Exception Type: RuntimeError, Error Details: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying), Traceback: Traceback (most recent call last):
+  File "/workspace/model_spec_midtraining/safety-tooling/safetytooling/apis/inference/openrouter.py", line 325, in __call__
+    raise RuntimeError(f"Empty response from {model_id} (common for openrouter so retrying)")
+RuntimeError: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying)
+.
+Retrying now. (Attempt 1)
+2026-08-28 23:19:34 [WARNING] (safetytooling.apis.inference.openrouter) Encountered API error: Exception Type: RuntimeError, Error Details: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying), Traceback: Traceback (most recent call last):
+  File "/workspace/model_spec_midtraining/safety-tooling/safetytooling/apis/inference/openrouter.py", line 325, in __call__
+    raise RuntimeError(f"Empty response from {model_id} (common for openrouter so retrying)")
+RuntimeError: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying)
+.
+Retrying now. (Attempt 0)
+2026-08-28 23:20:34 [WARNING] (safetytooling.apis.inference.openrouter) Encountered API error: Exception Type: RuntimeError, Error Details: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying), Traceback: Traceback (most recent call last):
+  File "/workspace/model_spec_midtraining/safety-tooling/safetytooling/apis/inference/openrouter.py", line 325, in __call__
+    raise RuntimeError(f"Empty response from {model_id} (common for openrouter so retrying)")
+RuntimeError: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying)
+.
+Retrying now. (Attempt 0)
+2026-08-28 23:22:23 [WARNING] (safetytooling.apis.inference.openrouter) Encountered API error: Exception Type: RuntimeError, Error Details: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying), Traceback: Traceback (most recent call last):
+  File "/workspace/model_spec_midtraining/safety-tooling/safetytooling/apis/inference/openrouter.py", line 325, in __call__
+    raise RuntimeError(f"Empty response from {model_id} (common for openrouter so retrying)")
+RuntimeError: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying)
+.
+Retrying now. (Attempt 0)
+2026-08-28 23:22:52 [WARNING] (safetytooling.apis.inference.openrouter) Encountered API error: Exception Type: RuntimeError, Error Details: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying), Traceback: Traceback (most recent call last):
+  File "/workspace/model_spec_midtraining/safety-tooling/safetytooling/apis/inference/openrouter.py", line 325, in __call__
+    raise RuntimeError(f"Empty response from {model_id} (common for openrouter so retrying)")
+RuntimeError: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying)
+.
+Retrying now. (Attempt 0)
+2026-08-28 23:26:22 [WARNING] (safetytooling.apis.inference.openrouter) Encountered API error: Exception Type: RuntimeError, Error Details: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying), Traceback: Traceback (most recent call last):
+  File "/workspace/model_spec_midtraining/safety-tooling/safetytooling/apis/inference/openrouter.py", line 325, in __call__
+    raise RuntimeError(f"Empty response from {model_id} (common for openrouter so retrying)")
+RuntimeError: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying)
+.
+Retrying now. (Attempt 0)
+2026-08-28 23:28:28 [WARNING] (safetytooling.apis.inference.openrouter) Encountered API error: Exception Type: RuntimeError, Error Details: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying), Traceback: Traceback (most recent call last):
+  File "/workspace/model_spec_midtraining/safety-tooling/safetytooling/apis/inference/openrouter.py", line 325, in __call__
+    raise RuntimeError(f"Empty response from {model_id} (common for openrouter so retrying)")
+RuntimeError: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying)
+.
+Retrying now. (Attempt 0)
+2026-08-28 23:30:36 [WARNING] (safetytooling.apis.inference.openrouter) Encountered API error: Exception Type: RuntimeError, Error Details: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying), Traceback: Traceback (most recent call last):
+  File "/workspace/model_spec_midtraining/safety-tooling/safetytooling/apis/inference/openrouter.py", line 325, in __call__
+    raise RuntimeError(f"Empty response from {model_id} (common for openrouter so retrying)")
+RuntimeError: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying)
+.
+Retrying now. (Attempt 0)
+2026-08-28 23:30:37 [WARNING] (safetytooling.apis.inference.openrouter) Encountered API error: Exception Type: RuntimeError, Error Details: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying), Traceback: Traceback (most recent call last):
+  File "/workspace/model_spec_midtraining/safety-tooling/safetytooling/apis/inference/openrouter.py", line 325, in __call__
+    raise RuntimeError(f"Empty response from {model_id} (common for openrouter so retrying)")
+RuntimeError: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying)
+.
+Retrying now. (Attempt 0)
+2026-08-28 23:33:38 [WARNING] (safetytooling.apis.inference.openrouter) Encountered API error: Exception Type: RuntimeError, Error Details: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying), Traceback: Traceback (most recent call last):
+  File "/workspace/model_spec_midtraining/safety-tooling/safetytooling/apis/inference/openrouter.py", line 325, in __call__
+    raise RuntimeError(f"Empty response from {model_id} (common for openrouter so retrying)")
+RuntimeError: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying)
+.
+Retrying now. (Attempt 0)

midtrain/qwen/deepseek-v4-pro/sycophancy/logs/run-2026-08-28-23-37-48-829645Z.log

@@ -0,0 +1,10 @@
+2026-08-28 23:37:48 [INFO] (safetytooling.utils.experiment_utils) Experiment configuration: DataGeneratorConfig(output_dir=PosixPath('data/midtrain/sycophancy_msm_deepseek-v4-pro_qwen'), enable_cache=False, cache_dir=None, use_redis=False, enable_prompt_history=False, prompt_history_dir=None, max_mem_usage_mb=5000, log_to_file=True, openai_fraction_rate_limit=0.5, openai_num_threads=50, openai_base_url=None, gemini_num_threads=100, openai_s2s_num_threads=40, gpt4o_s2s_rpm_cap=10, gemini_recitation_rate_check_volume=100, gemini_recitation_rate_threshold=0.5, anthropic_num_threads=40, gray_swan_num_threads=40, huggingface_num_threads=100, together_num_threads=40, openrouter_num_threads=40, vllm_num_threads=8, use_vllm_if_model_not_found=False, vllm_base_url='http://localhost:8000/v1/chat/completions', openai_tag='OPENAI_API_KEY', anthropic_tag='ANTHROPIC_API_KEY', openrouter_tag='OPENROUTER_API_KEY', print_prompt_and_response=False, datetime_str='2026-08-28-23-37-48-829645Z', seed=42, principle_name='sycophancy', dataset_name='sycophancy_msm_deepseek-v4-pro_qwen', spec_file_name='sycophancy', model_name='Qwen', provider_name='Alibaba', spec_type='default', model_id='deepseek/deepseek-v4-pro', max_output_tokens=64000, temperature=1.0, n_domains=5, n_doc_ideas=16, n_doc_types=16, specified_domains=[], max_concurrent_requests=30, use_batch_api=False, batch_timeout_minutes=180, anthropic_batch_tag='ANTHROPIC_BATCH_API_KEY', tokenizer_name='meta-llama/Llama-3.1-8B', exact_token_count=False, preview=False, data_dir=PosixPath('data/gen_synth_docs/sycophancy_msm_deepseek-v4-pro_qwen'))
+2026-08-28 23:37:48 [WARNING] (safetytooling.utils.utils) OPENAI_API_KEY not found in environment, OpenAI API will not be available
+2026-08-28 23:37:48 [WARNING] (safetytooling.utils.utils) ANTHROPIC_API_KEY not found in environment, Anthropic API will not be available
+2026-08-28 23:37:48 [WARNING] (safetytooling.utils.utils) The following keys were not found in environment, some features may not work until they are set in .env: ['DEEPSEEK_API_KEY']
+2026-08-28 23:40:12 [WARNING] (safetytooling.apis.inference.openrouter) Encountered API error: Exception Type: RuntimeError, Error Details: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying), Traceback: Traceback (most recent call last):
+  File "/workspace/model_spec_midtraining/safety-tooling/safetytooling/apis/inference/openrouter.py", line 325, in __call__
+    raise RuntimeError(f"Empty response from {model_id} (common for openrouter so retrying)")
+RuntimeError: Empty response from deepseek/deepseek-v4-pro (common for openrouter so retrying)
+.
+Retrying now. (Attempt 0)

midtrain/qwen/deepseek-v4-pro/sycophancy/summary.json

@@ -0,0 +1,28 @@
+{
+  "config": {
+    "principle_name": "sycophancy",
+    "dataset_name": "sycophancy_msm_deepseek-v4-pro_qwen",
+    "spec_file_name": "sycophancy",
+    "model_name": "Qwen",
+    "provider_name": "Alibaba",
+    "spec_type": "default",
+    "model_id": "deepseek/deepseek-v4-pro",
+    "max_output_tokens": 64000,
+    "temperature": 1.0,
+    "n_domains": 5,
+    "n_doc_types": 16,
+    "n_doc_ideas": 16,
+    "specified_domains": [],
+    "use_batch_api": false,
+    "tokenizer_name": "meta-llama/Llama-3.1-8B"
+  },
+  "stats": {
+    "n_domains": 4,
+    "n_subdomains": 20,
+    "n_assertions": 81,
+    "n_documents": 4913,
+    "total_tokens": 5981803,
+    "tokens_per_document": 1224,
+    "max_token_length": 2859
+  }
+}