stage1 v3pf qwen3_5_9b

.gitattributes

@@ -44,3 +44,4 @@ qwen3_5_0_8b_stage2_v3/vllm_compat/tokenizer.json filter=lfs diff=lfs merge=lfs
 qwen3_5_2b_stage2_v3/vllm_compat/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 qwen3_5_9b_stage2_v3/vllm_compat/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 stage1_v3_qwen3_6_27b/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+stage1_v3pf_qwen3_5_9b/tokenizer.json filter=lfs diff=lfs merge=lfs -text

stage1_v3pf_qwen3_5_9b/chat_template.jinja

@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- else %}
+        {{- '<think>\n' }}
+    {%- endif %}
+{%- endif %}

stage1_v3pf_qwen3_5_9b/loss_curve.csv

@@ -0,0 +1,122 @@
+step,loss,lr,tok_per_s
+0,4.040378,1.000e-05,325.0
+5,4.111278,6.000e-05,10930.3
+10,3.869833,1.100e-04,11729.9
+15,3.834497,1.600e-04,8877.6
+20,3.855317,2.100e-04,11108.2
+25,3.887257,2.600e-04,11784.7
+30,3.779253,3.100e-04,11828.4
+35,3.612720,3.600e-04,11847.5
+40,3.425594,4.100e-04,11831.8
+45,3.322668,4.600e-04,11849.1
+50,3.133181,5.100e-04,11845.3
+55,2.865881,5.600e-04,11850.6
+60,2.771411,6.100e-04,11843.7
+65,2.899855,6.600e-04,8523.1
+70,2.869309,7.100e-04,11846.5
+75,2.785730,7.600e-04,11850.0
+80,2.703507,8.100e-04,11848.0
+85,2.104745,8.600e-04,11861.8
+90,2.139479,9.100e-04,11849.3
+95,2.476314,9.600e-04,11866.9
+100,2.215527,1.000e-03,11853.4
+105,2.182172,9.998e-04,11855.8
+110,2.225345,9.990e-04,11854.8
+115,2.236989,9.978e-04,11841.7
+120,1.996411,9.961e-04,11854.6
+125,1.689849,9.938e-04,11861.2
+130,1.718480,9.911e-04,11857.9
+135,1.570013,9.880e-04,11862.8
+140,1.777773,9.843e-04,11854.1
+145,1.606067,9.801e-04,11863.0
+150,1.449982,9.755e-04,11854.9
+155,1.395623,9.704e-04,11845.3
+160,1.436482,9.649e-04,11854.8
+165,1.526646,9.589e-04,11859.0
+170,1.482958,9.524e-04,11858.3
+175,1.394650,9.455e-04,11860.4
+180,1.379590,9.382e-04,11857.6
+185,1.274665,9.304e-04,11865.2
+190,1.492298,9.222e-04,11855.5
+195,1.491493,9.135e-04,11861.5
+200,1.479487,9.045e-04,11106.1
+205,1.335710,8.951e-04,11861.0
+210,1.544783,8.853e-04,11859.1
+215,1.456074,8.751e-04,8680.3
+220,1.472516,8.645e-04,11836.3
+225,1.375135,8.536e-04,11847.9
+230,1.400707,8.423e-04,11853.0
+235,1.376647,8.307e-04,11853.1
+240,1.212857,8.187e-04,11081.5
+245,1.478824,8.065e-04,11841.7
+250,1.603603,7.939e-04,11843.9
+255,1.500814,7.810e-04,11840.0
+260,1.409045,7.679e-04,11840.7
+265,1.282132,7.545e-04,11850.1
+270,1.177373,7.409e-04,11845.6
+275,1.160959,7.270e-04,11850.9
+280,1.132518,7.129e-04,11851.6
+285,1.088371,6.986e-04,11845.0
+290,1.430400,6.841e-04,11842.8
+295,1.293683,6.694e-04,11850.8
+300,1.219701,6.545e-04,9854.4
+305,1.453785,6.395e-04,11802.2
+310,1.433822,6.243e-04,11843.2
+315,1.371349,6.091e-04,11854.1
+320,1.263873,5.937e-04,11845.5
+325,1.292316,5.782e-04,11852.1
+330,1.162319,5.627e-04,11852.5
+335,1.161981,5.471e-04,11851.1
+340,1.325282,5.314e-04,11848.0
+345,1.246090,5.157e-04,11847.3
+350,1.209881,5.000e-04,11848.5
+355,1.163959,4.843e-04,10070.7
+360,1.396391,4.686e-04,11848.4
+365,1.464400,4.529e-04,11848.6
+370,1.394919,4.373e-04,11846.4
+375,1.259350,4.218e-04,11849.1
+380,1.083231,4.063e-04,11848.7
+385,1.189685,3.909e-04,11842.5
+390,1.233931,3.757e-04,11843.6
+395,1.344021,3.605e-04,11850.7
+400,1.177833,3.455e-04,11851.4
+405,1.276042,3.306e-04,11848.7
+410,1.288682,3.159e-04,11844.4
+415,1.210146,3.014e-04,11851.9
+420,1.263751,2.871e-04,11844.9
+425,1.115266,2.730e-04,11849.6
+430,1.129994,2.591e-04,11840.8
+435,1.280874,2.455e-04,11843.7
+440,1.150072,2.321e-04,11851.3
+445,1.272855,2.190e-04,11856.1
+450,1.116954,2.061e-04,10048.0
+455,1.230477,1.935e-04,11825.0
+460,1.225651,1.813e-04,11816.6
+465,1.264480,1.693e-04,11831.6
+470,1.217883,1.577e-04,11826.0
+475,1.236631,1.464e-04,11838.2
+480,1.195988,1.355e-04,11834.4
+485,1.227494,1.249e-04,11840.4
+490,1.237088,1.147e-04,11833.3
+495,1.195743,1.049e-04,11851.1
+500,1.212842,9.549e-05,11844.0
+505,1.217877,8.646e-05,11851.8
+510,1.207031,7.784e-05,11848.4
+515,1.197403,6.963e-05,11847.0
+520,1.217326,6.185e-05,11850.3
+525,1.271957,5.450e-05,11854.0
+530,1.264466,4.759e-05,10826.6
+535,1.244463,4.112e-05,11853.2
+540,1.302761,3.511e-05,11316.5
+545,1.265297,2.956e-05,11846.7
+550,1.175814,2.447e-05,11845.9
+555,1.151829,1.985e-05,11850.5
+560,1.158942,1.571e-05,11842.8
+565,1.216073,1.204e-05,11851.9
+570,1.345215,8.856e-06,11843.6
+575,1.251737,6.156e-06,11854.2
+580,1.285080,3.943e-06,11838.9
+585,1.421452,2.219e-06,11841.8
+590,1.351571,9.866e-07,11838.7
+595,1.389049,2.467e-07,11843.2
+599,1.335247,9.870e-09,11849.9

stage1_v3pf_qwen3_5_9b/stage1_summary.json

@@ -0,0 +1,49 @@
+{
+  "model": "Qwen/Qwen3.5-9B",
+  "probe": "probe_results/qwen3_5_9b_v3pf.json",
+  "steps": 600,
+  "lr": 0.001,
+  "seq_len": 4096,
+  "batch_size": 1,
+  "d_idx": 16,
+  "elapsed_min": 3.70496234225,
+  "final_per_head_kl": {
+    "3": [
+      1.5307061672210693,
+      1.4760429859161377
+    ],
+    "7": [
+      1.4086997509002686
+    ],
+    "11": [
+      1.6310946941375732
+    ],
+    "15": [
+      1.5815995931625366
+    ],
+    "19": [
+      1.0981472730636597,
+      1.2471306324005127,
+      1.3031089305877686,
+      1.4784331321716309,
+      1.6598527431488037,
+      1.2980384826660156
+    ],
+    "23": [
+      1.1785054206848145,
+      1.1001001596450806,
+      1.0274646282196045,
+      1.2863967418670654
+    ],
+    "27": [
+      1.225365400314331,
+      1.213803768157959,
+      1.4813830852508545
+    ],
+    "31": [
+      0.7547898292541504
+    ]
+  },
+  "final_mean_kl": 1.3147717588826229,
+  "final_max_kl": 1.6598527431488037
+}

stage1_v3pf_qwen3_5_9b/state_dict.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ebd3a56711d914e594052d1a59a9425b73a531a2577c835dc5353544cec0e03
+size 325045

stage1_v3pf_qwen3_5_9b/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:06b9509352d2af50381ab2247e083b80d32d5c0aba91c272ca9ff729b6a0e523
+size 19989325

stage1_v3pf_qwen3_5_9b/tokenizer_config.json

@@ -0,0 +1,32 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "local_files_only": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}