Model save

README.md

@@ -1,5 +1,5 @@
 ---
-base_model: Qwen/Qwen2.5-7B
+base_model: Qwen/Qwen2.5-0.5B
 library_name: transformers
 model_name: Qwen-2.5-7B-Simple-RL
 tags:
@@ -11,7 +11,7 @@ licence: license
 
 # Model Card for Qwen-2.5-7B-Simple-RL
 
-This model is a fine-tuned version of [Qwen/Qwen2.5-7B](https://huggingface.co/Qwen/Qwen2.5-7B).
+This model is a fine-tuned version of [Qwen/Qwen2.5-0.5B](https://huggingface.co/Qwen/Qwen2.5-0.5B).
 It has been trained using [TRL](https://github.com/huggingface/trl).
 
 ## Quick start
@@ -27,7 +27,7 @@ print(output["generated_text"])
 
 ## Training procedure
 
-[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/jzhang-x-01/huggingface/runs/38wymkvl) 
+[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/jzhang-x-01/huggingface/runs/gj9p99u7) 
 
 
 This model was trained with GRPO, a method introduced in [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://huggingface.co/papers/2402.03300).

all_results.json

@@ -1,8 +1,8 @@
 {
     "total_flos": 0.0,
-    "train_loss": 0.0010652160639474098,
-    "train_runtime": 7821.2028,
-    "train_samples": 7500,
-    "train_samples_per_second": 0.959,
-    "train_steps_per_second": 0.007
+    "train_loss": 2.8635031776502727e-05,
+    "train_runtime": 365.6249,
+    "train_samples": 750,
+    "train_samples_per_second": 2.051,
+    "train_steps_per_second": 0.014
 }

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "Qwen/Qwen2.5-7B",
+  "_name_or_path": "Qwen/Qwen2.5-0.5B",
   "architectures": [
     "Qwen2ForCausalLM"
   ],
@@ -7,24 +7,24 @@
   "bos_token_id": 151643,
   "eos_token_id": 151643,
   "hidden_act": "silu",
-  "hidden_size": 3584,
+  "hidden_size": 896,
   "initializer_range": 0.02,
-  "intermediate_size": 18944,
-  "max_position_embeddings": 131072,
-  "max_window_layers": 28,
+  "intermediate_size": 4864,
+  "max_position_embeddings": 32768,
+  "max_window_layers": 24,
   "model_type": "qwen2",
-  "num_attention_heads": 28,
-  "num_hidden_layers": 28,
-  "num_key_value_heads": 4,
+  "num_attention_heads": 14,
+  "num_hidden_layers": 24,
+  "num_key_value_heads": 2,
   "rms_norm_eps": 1e-06,
   "rope_scaling": null,
   "rope_theta": 1000000.0,
-  "sliding_window": 131072,
-  "tie_word_embeddings": false,
+  "sliding_window": 32768,
+  "tie_word_embeddings": true,
   "torch_dtype": "bfloat16",
   "transformers_version": "4.49.0",
   "use_cache": false,
   "use_mrope": false,
   "use_sliding_window": false,
-  "vocab_size": 152064
+  "vocab_size": 151936
 }

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5645c1de0b30d0217ecac6927108da17c19791fe8675832336c0dd33c4102e06
+size 988097824

train_results.json

@@ -1,8 +1,8 @@
 {
     "total_flos": 0.0,
-    "train_loss": 0.0010652160639474098,
-    "train_runtime": 7821.2028,
-    "train_samples": 7500,
-    "train_samples_per_second": 0.959,
-    "train_steps_per_second": 0.007
+    "train_loss": 2.8635031776502727e-05,
+    "train_runtime": 365.6249,
+    "train_samples": 750,
+    "train_samples_per_second": 2.051,
+    "train_steps_per_second": 0.014
 }

trainer_state.json

@@ -1,187 +1,51 @@
 {
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 0.9893390191897654,
+  "epoch": 0.851063829787234,
   "eval_steps": 100,
-  "global_step": 58,
+  "global_step": 5,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
-      "completion_length": 473.94757080078125,
-      "epoch": 0.017057569296375266,
-      "grad_norm": 4.064376038455483,
+      "completion_length": 558.3560485839844,
+      "epoch": 0.1702127659574468,
+      "grad_norm": 0.202052697018779,
       "kl": 0.0,
-      "learning_rate": 5e-07,
+      "learning_rate": 3e-06,
       "loss": 0.0,
-      "reward": 0.1729910783469677,
-      "reward_std": 0.27430381812155247,
-      "rewards/accuracy_reward": 0.1729910783469677,
+      "reward": 0.017857143888249993,
+      "reward_std": 0.04515197407454252,
+      "rewards/accuracy_reward": 0.017857143888249993,
       "rewards/format_reward": 0.0,
       "step": 1
     },
     {
-      "completion_length": 485.79745292663574,
-      "epoch": 0.08528784648187633,
-      "grad_norm": 1.1351739582457168,
-      "kl": 0.0002506077289581299,
-      "learning_rate": 2.5e-06,
+      "completion_length": 560.4059982299805,
+      "epoch": 0.851063829787234,
+      "grad_norm": 0.23003411642703778,
+      "kl": 0.0008947253227233887,
+      "learning_rate": 0.0,
       "loss": 0.0,
-      "reward": 0.13504464970901608,
-      "reward_std": 0.21868262230418622,
-      "rewards/accuracy_reward": 0.1347656319849193,
-      "rewards/format_reward": 0.00027901786961592734,
-      "step": 5
-    },
-    {
-      "completion_length": 568.8616348266602,
-      "epoch": 0.17057569296375266,
-      "grad_norm": 0.5575129470407404,
-      "kl": 0.015078353881835937,
-      "learning_rate": 2.956412726139078e-06,
-      "loss": 0.0006,
-      "reward": 0.3587053745985031,
-      "reward_std": 0.3362843289971352,
-      "rewards/accuracy_reward": 0.35848215967416763,
-      "rewards/format_reward": 0.00022321429569274187,
-      "step": 10
-    },
-    {
-      "completion_length": 584.8319480895996,
-      "epoch": 0.255863539445629,
-      "grad_norm": 0.2277859906277476,
-      "kl": 0.02491455078125,
-      "learning_rate": 2.7836719084521715e-06,
-      "loss": 0.001,
-      "reward": 0.638169676065445,
-      "reward_std": 0.24866362921893598,
-      "rewards/accuracy_reward": 0.638169676065445,
+      "reward": 0.025948662223527208,
+      "reward_std": 0.06277249014237896,
+      "rewards/accuracy_reward": 0.025948662223527208,
       "rewards/format_reward": 0.0,
-      "step": 15
-    },
-    {
-      "completion_length": 535.3618560791016,
-      "epoch": 0.3411513859275053,
-      "grad_norm": 0.23887369640814476,
-      "kl": 0.0245208740234375,
-      "learning_rate": 2.4946839873611927e-06,
-      "loss": 0.001,
-      "reward": 0.6774553880095482,
-      "reward_std": 0.2348104739561677,
-      "rewards/accuracy_reward": 0.6767857447266579,
-      "rewards/format_reward": 0.0006696428870782256,
-      "step": 20
-    },
-    {
-      "completion_length": 542.0183288574219,
-      "epoch": 0.42643923240938164,
-      "grad_norm": 0.3014825446189458,
-      "kl": 0.02587432861328125,
-      "learning_rate": 2.1156192081791355e-06,
-      "loss": 0.001,
-      "reward": 0.6584821730852127,
-      "reward_std": 0.22855298295617105,
-      "rewards/accuracy_reward": 0.657589316368103,
-      "rewards/format_reward": 0.0008928571827709675,
-      "step": 25
-    },
-    {
-      "completion_length": 536.8770370483398,
-      "epoch": 0.511727078891258,
-      "grad_norm": 0.21529686209177487,
-      "kl": 0.0219482421875,
-      "learning_rate": 1.6808050203829845e-06,
-      "loss": 0.0009,
-      "reward": 0.6495536014437675,
-      "reward_std": 0.2183901211246848,
-      "rewards/accuracy_reward": 0.6488839566707612,
-      "rewards/format_reward": 0.0006696428870782256,
-      "step": 30
-    },
-    {
-      "completion_length": 528.6895355224609,
-      "epoch": 0.5970149253731343,
-      "grad_norm": 0.16077481890595569,
-      "kl": 0.0267181396484375,
-      "learning_rate": 1.2296174432791415e-06,
-      "loss": 0.0011,
-      "reward": 0.6448661029338837,
-      "reward_std": 0.2145981529727578,
-      "rewards/accuracy_reward": 0.6441964589059352,
-      "rewards/format_reward": 0.0006696428870782256,
-      "step": 35
-    },
-    {
-      "completion_length": 514.0154296875,
-      "epoch": 0.6823027718550106,
-      "grad_norm": 0.8368130250058291,
-      "kl": 0.02160186767578125,
-      "learning_rate": 8.029152419343472e-07,
-      "loss": 0.0009,
-      "reward": 0.6649553939700127,
-      "reward_std": 0.20347560551017524,
-      "rewards/accuracy_reward": 0.6647321790456772,
-      "rewards/format_reward": 0.00022321429569274187,
-      "step": 40
-    },
-    {
-      "completion_length": 528.3629722595215,
-      "epoch": 0.767590618336887,
-      "grad_norm": 0.3144943865256562,
-      "kl": 0.0201324462890625,
-      "learning_rate": 4.3933982822017883e-07,
-      "loss": 0.0008,
-      "reward": 0.6553571708500385,
-      "reward_std": 0.21330095445737243,
-      "rewards/accuracy_reward": 0.6542410992085934,
-      "rewards/format_reward": 0.0011160714784637094,
-      "step": 45
-    },
-    {
-      "completion_length": 526.5335037231446,
-      "epoch": 0.8528784648187633,
-      "grad_norm": 2.72821016556171,
-      "kl": 0.027325439453125,
-      "learning_rate": 1.718159615201853e-07,
-      "loss": 0.0011,
-      "reward": 0.6439732402563095,
-      "reward_std": 0.22680639270693065,
-      "rewards/accuracy_reward": 0.6437500268220901,
-      "rewards/format_reward": 0.00022321429569274187,
-      "step": 50
-    },
-    {
-      "completion_length": 523.5705612182617,
-      "epoch": 0.9381663113006397,
-      "grad_norm": 0.41062600235384844,
-      "kl": 0.0195343017578125,
-      "learning_rate": 2.4570139579284723e-08,
-      "loss": 0.0008,
-      "reward": 0.6685268178582191,
-      "reward_std": 0.23179882913827896,
-      "rewards/accuracy_reward": 0.6676339596509934,
-      "rewards/format_reward": 0.0008928571827709675,
-      "step": 55
+      "step": 5
     },
     {
-      "completion_length": 517.845261891683,
-      "epoch": 0.9893390191897654,
-      "kl": 0.021814982096354168,
-      "reward": 0.6592262188593546,
-      "reward_std": 0.20016193420936665,
-      "rewards/accuracy_reward": 0.6588541939854622,
-      "rewards/format_reward": 0.0003720238261545698,
-      "step": 58,
+      "epoch": 0.851063829787234,
+      "step": 5,
       "total_flos": 0.0,
-      "train_loss": 0.0010652160639474098,
-      "train_runtime": 7821.2028,
-      "train_samples_per_second": 0.959,
-      "train_steps_per_second": 0.007
+      "train_loss": 2.8635031776502727e-05,
+      "train_runtime": 365.6249,
+      "train_samples_per_second": 2.051,
+      "train_steps_per_second": 0.014
     }
   ],
   "logging_steps": 5,
-  "max_steps": 58,
+  "max_steps": 5,
   "num_input_tokens_seen": 0,
   "num_train_epochs": 1,
   "save_steps": 500,

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e9c97f93968e70f80d79d3367db73181a5a13edc4aad2638a46a6504eb5664cf
-size 7928
+oid sha256:0331e1487175abe417a3ec2bca598d4d79190d4f1202fbdf1338d0683e148e6b
+size 8056