Upload promoted hard recall micro boost adapter

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,26 @@
+---
+base_model: Qwen/Qwen3-1.7B
+license: mit
+library_name: peft
+pipeline_tag: text-generation
+---
+
+# AtomicVision Hard Recall Micro Boost LoRA
+
+This adapter materializes the promoted `checkpoint-1` checkpoint from the
+targeted hard recall micro-repair continuation.
+
+## Parent
+- Base adapter: [prodigyhuh/atomicvision-medium-fidelity-boost-lora](https://huggingface.co/prodigyhuh/atomicvision-medium-fidelity-boost-lora)
+- Source HF job: [69ed269fd70108f37acdef6d](https://huggingface.co/jobs/prodigyhuh/69ed269fd70108f37acdef6d)
+- Source commit: `3838f9048bce4c6bc81e57f5c0dab00980c7fa08`
+
+## Held-out strict eval (`seed_start=10000`, `episodes=32`)
+
+| Difficulty | Reward | F1 | MAE | Strict | Normalized | Done | Submit |
+| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| medium | 4.5065 | 0.7891 | 0.02712 | 1.00 | 1.00 | 1.00 | 1.00 |
+| hard | 4.7148 | 0.8207 | 0.02552 | 1.00 | 1.00 | 1.00 | 1.00 |
+
+This run preserves perfect strict execution and slightly improves the hard slice
+over the previous best published adapter without regressing medium.

adapter_config.json

@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3-1.7B",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.05,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "k_proj",
+    "down_proj",
+    "gate_proj",
+    "o_proj",
+    "q_proj",
+    "up_proj",
+    "v_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:745af6dec82baa429af53c2feba7c5832ef2f990bcc8d97f680822e5ea33f110
+size 69782384

chat_template.jinja

@@ -0,0 +1,93 @@
+{#- Training variant of the Qwen3 chat template (see qwen3.jinja for the original).
+    Modifications vs the original:
+    - {%- if '</think>' in content %} → {%- if '<think>' in content and '</think>' in content %}
+      Always check for both tags to avoid edge cases where the model generates only one tag.
+    - Removed the loop.index0 > ns.last_query_index conditional; always include thinking block.
+      This makes the template prefix-preserving for the [user, assistant] → [user, assistant, tool] transition.
+    - Added {% generation %} / {% endgeneration %} around assistant message output to support
+      assistant-only loss masking in SFT training.
+-#}
+{%- if tools %}
+    {{- '<|im_start|>system\n' }}
+    {%- if messages[0].role == 'system' %}
+        {{- messages[0].content + '\n\n' }}
+    {%- endif %}
+    {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {{- '<|im_start|>system\n' + messages[0].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" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
+        {%- set ns.multi_step_tool = false %}
+        {%- set ns.last_query_index = index %}
+    {%- endif %}
+{%- endfor %}
+{%- for message in messages %}
+    {%- if message.content is string %}
+        {%- set content = message.content %}
+    {%- else %}
+        {%- set content = '' %}
+    {%- endif %}
+    {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
+        {{- '<|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 and '</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 %}
+        {{- '<|im_start|>' + message.role + '\n' }}
+        {%- generation %}
+        {{- '<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
+        {%- if message.tool_calls %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if (loop.first and content) or (not loop.first) %}
+                    {{- '\n' }}
+                {%- endif %}
+                {%- if tool_call.function %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {{- '<tool_call>\n{"name": "' }}
+                {{- tool_call.name }}
+                {{- '", "arguments": ' }}
+                {%- if tool_call.arguments is string %}
+                    {{- tool_call.arguments }}
+                {%- else %}
+                    {{- tool_call.arguments | tojson }}
+                {%- endif %}
+                {{- '}\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+        {%- endgeneration %}
+    {%- elif message.role == "tool" %}
+        {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- 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' }}
+    {%- endif %}
+{%- endif %}

heldout_eval.json

@@ -0,0 +1,134 @@
+{
+  "base_model": "Qwen/Qwen3-1.7B",
+  "adapter": "/tmp/atomicvision_publish_runner/output/train/checkpoint-1",
+  "episodes_per_difficulty": 32,
+  "seed_start": 10000,
+  "seed_policy": {
+    "sft_train": {
+      "start": 1000,
+      "stop": 4000
+    },
+    "grpo_train": {
+      "start": 4000,
+      "stop": 8000
+    },
+    "heldout_eval": {
+      "start": 10000,
+      "stop": 11000
+    }
+  },
+  "heldout_seed_enforced": true,
+  "max_tool_steps": 3,
+  "max_new_tokens": 180,
+  "modes": [
+    "strict"
+  ],
+  "results": {
+    "medium": {
+      "baseline_prior_submit": {
+        "episodes": 32,
+        "mean_reward": 4.65651721875,
+        "mean_f1": 0.80580365625,
+        "mean_mae": 0.0244615625,
+        "mean_steps": 2.0,
+        "mean_scan_cost": 1.5,
+        "done_rate": 1.0,
+        "tool_failure_rate": 0.0,
+        "mean_repeated_tool_calls": 0.0,
+        "strict_tool_call_pass_rate": 1.0,
+        "normalized_tool_call_pass_rate": 1.0,
+        "normalized_tool_call_repair_rate": 0.0,
+        "first_action_valid_rate": 1.0,
+        "first_action_ask_prior_rate": 1.0,
+        "submit_action_rate": 1.0,
+        "mean_identity_reward": 3.2232142812499998,
+        "mean_concentration_reward": 2.23155896875,
+        "mean_confidence_reward": 0.270494,
+        "mean_false_positive_penalty": -0.1875,
+        "mean_missed_defect_penalty": -0.28125,
+        "mean_timeout_penalty": 0.0,
+        "mean_outcome_reward_total": 5.72526725,
+        "mean_penalty_total": -1.06875
+      },
+      "strict_adapter": {
+        "episodes": 32,
+        "mean_reward": 4.50648265625,
+        "mean_f1": 0.789137,
+        "mean_mae": 0.027124218749999998,
+        "mean_steps": 2.0,
+        "mean_scan_cost": 1.5,
+        "done_rate": 1.0,
+        "tool_failure_rate": 0.0,
+        "mean_repeated_tool_calls": 0.0,
+        "strict_tool_call_pass_rate": 1.0,
+        "normalized_tool_call_pass_rate": 1.0,
+        "normalized_tool_call_repair_rate": 0.0,
+        "first_action_valid_rate": 1.0,
+        "first_action_ask_prior_rate": 1.0,
+        "submit_action_rate": 1.0,
+        "mean_identity_reward": 3.156547625,
+        "mean_concentration_reward": 2.16269634375,
+        "mean_confidence_reward": 0.29348875,
+        "mean_false_positive_penalty": -0.1875,
+        "mean_missed_defect_penalty": -0.31875,
+        "mean_timeout_penalty": 0.0,
+        "mean_outcome_reward_total": 5.61273271875,
+        "mean_penalty_total": -1.10625
+      },
+      "strict_failures": []
+    },
+    "hard": {
+      "baseline_prior_submit": {
+        "episodes": 32,
+        "mean_reward": 5.01651990625,
+        "mean_f1": 0.85153328125,
+        "mean_mae": 0.02220903125,
+        "mean_steps": 2.0,
+        "mean_scan_cost": 1.5,
+        "done_rate": 1.0,
+        "tool_failure_rate": 0.0,
+        "mean_repeated_tool_calls": 0.0,
+        "strict_tool_call_pass_rate": 1.0,
+        "normalized_tool_call_pass_rate": 1.0,
+        "normalized_tool_call_repair_rate": 0.0,
+        "first_action_valid_rate": 1.0,
+        "first_action_ask_prior_rate": 1.0,
+        "submit_action_rate": 1.0,
+        "mean_identity_reward": 3.40613278125,
+        "mean_concentration_reward": 2.3444645625,
+        "mean_confidence_reward": 0.53779759375,
+        "mean_false_positive_penalty": -0.109375,
+        "mean_missed_defect_penalty": -0.5625,
+        "mean_timeout_penalty": 0.0,
+        "mean_outcome_reward_total": 6.2883949375,
+        "mean_penalty_total": -1.2718749999999999
+      },
+      "strict_adapter": {
+        "episodes": 32,
+        "mean_reward": 4.714775875,
+        "mean_f1": 0.8206800937500001,
+        "mean_mae": 0.02552296875,
+        "mean_steps": 2.0,
+        "mean_scan_cost": 1.5,
+        "done_rate": 1.0,
+        "tool_failure_rate": 0.0,
+        "mean_repeated_tool_calls": 0.0,
+        "strict_tool_call_pass_rate": 1.0,
+        "normalized_tool_call_pass_rate": 1.0,
+        "normalized_tool_call_repair_rate": 0.0,
+        "first_action_valid_rate": 1.0,
+        "first_action_ask_prior_rate": 1.0,
+        "submit_action_rate": 1.0,
+        "mean_identity_reward": 3.282720125,
+        "mean_concentration_reward": 2.243760375,
+        "mean_confidence_reward": 0.5257955,
+        "mean_false_positive_penalty": -0.09375,
+        "mean_missed_defect_penalty": -0.64375,
+        "mean_timeout_penalty": 0.0,
+        "mean_outcome_reward_total": 6.052276,
+        "mean_penalty_total": -1.3375
+      },
+      "strict_failures": []
+    }
+  }
+}

promotion_summary.json

@@ -0,0 +1,81 @@
+{
+  "base_adapter": "/tmp/atomicvision_publish_runner/output/base_adapter",
+  "candidates": {
+    "base": {
+      "hard": {
+        "done": 1.0,
+        "f1": 0.81621578125,
+        "fail": 0.0,
+        "mae": 0.02591959375,
+        "normalized": 1.0,
+        "reward": 4.69171921875,
+        "strict": 1.0,
+        "submit": 1.0
+      },
+      "hard_f1_delta_vs_base": 0.0,
+      "hard_reward_delta_vs_base": 0.0,
+      "label": "base",
+      "medium": {
+        "done": 1.0,
+        "f1": 0.789137,
+        "fail": 0.0,
+        "mae": 0.027124218749999998,
+        "normalized": 1.0,
+        "reward": 4.50648265625,
+        "strict": 1.0,
+        "submit": 1.0
+      },
+      "medium_f1_delta_vs_base": 0.0,
+      "medium_reward_delta_vs_base": 0.0
+    },
+    "checkpoint-1": {
+      "hard": {
+        "done": 1.0,
+        "f1": 0.8206800937500001,
+        "fail": 0.0,
+        "mae": 0.02552296875,
+        "normalized": 1.0,
+        "reward": 4.714775875,
+        "strict": 1.0,
+        "submit": 1.0
+      },
+      "hard_f1_delta_vs_base": 0.004464312500000012,
+      "hard_reward_delta_vs_base": 0.02305665624999964,
+      "label": "checkpoint-1",
+      "medium": {
+        "done": 1.0,
+        "f1": 0.789137,
+        "fail": 0.0,
+        "mae": 0.027124218749999998,
+        "normalized": 1.0,
+        "reward": 4.50648265625,
+        "strict": 1.0,
+        "submit": 1.0
+      },
+      "medium_f1_delta_vs_base": 0.0,
+      "medium_reward_delta_vs_base": 0.0
+    }
+  },
+  "checkpoint_steps": [
+    1
+  ],
+  "dataset_counts": {
+    "submit_after_reference": 12,
+    "submit_prior": 4
+  },
+  "episodes_per_difficulty": 16,
+  "eval_difficulties": [
+    "medium",
+    "hard"
+  ],
+  "eval_episodes": 32,
+  "eval_seed_start": 10000,
+  "learning_rate": 1e-06,
+  "max_updates": 1,
+  "profile": "hard_recall_micro_repair",
+  "promotion_candidate": "checkpoint-1",
+  "seed_start": 3600,
+  "train_difficulties": [
+    "hard"
+  ]
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506
+size 11422650

tokenizer_config.json

@@ -0,0 +1,30 @@
+{
+  "add_prefix_space": false,
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "is_local": false,
+  "local_files_only": false,
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}