Upload best model checkpoint

metrics.json

@@ -1,30 +1,29 @@
 {
-    "rl_info/A2G": -0.00804260652512312,
-    "rl_info/entropy": 2.8796980381011963,
-    "rl_info/total_token": 440.0,
-    "rl_info/advantage_b4_norm": 961.4583129882812,
-    "rl_info/advantage_after_gnorm": -0.13672709465026855,
+    "rl_info/A2G": -0.07268727570772171,
+    "rl_info/entropy": 3.100099802017212,
+    "rl_info/total_token": 1794.0,
+    "rl_info/advantage_b4_norm": -329.8147277832031,
     "rl_info/kl_w_ref": 0.0,
-    "train/rl_loss": 0.8039727210998535,
-    "train/total_loss": 0.8039727210998535,
-    "xsum/rouge1": 0.1822673507912263,
-    "xsum/rouge2": 0.03530779605308503,
-    "xsum/rougeL": 0.1487245393010488,
-    "xsum/rougeLsum": 0.14915382086079024,
-    "xsum/bertscore_precision": 0.7578381817936898,
-    "xsum/bertscore_recall": 0.7217868343293666,
-    "xsum/bertscore_f1": 0.7386852816343308,
-    "eval_agg/avg_all_rougef": 0.12886337675153758,
-    "eval_agg/avg_all_bertf": 0.7386852816343308,
-    "eval_agg/avg_all": 0.4337743291929342,
-    "num_rl_rollout": 213,
+    "train/rl_loss": 7.268417835235596,
+    "train/total_loss": 7.268417835235596,
+    "xsum/rouge1": 0.16526458612022013,
+    "xsum/rouge2": 0.026317550263276528,
+    "xsum/rougeL": 0.13316665912096423,
+    "xsum/rougeLsum": 0.13414432999907605,
+    "xsum/bertscore_precision": 0.7271568326748551,
+    "xsum/bertscore_recall": 0.7270669051583575,
+    "xsum/bertscore_f1": 0.7266696648915777,
+    "eval_agg/avg_all_rougef": 0.11472328137588424,
+    "eval_agg/avg_all_bertf": 0.7266696648915777,
+    "eval_agg/avg_all": 0.420696473133731,
+    "num_rl_rollout": 5,
     "lm_epoch": 0,
     "rl_epoch": 0,
-    "step": 17000,
-    "total_data_token": 2906376,
-    "total_rl_token": 18963869,
+    "step": 1000,
+    "total_data_token": 63494,
+    "total_rl_token": 1974715,
     "total_lm_token": 0,
-    "total_token": 18963869,
-    "completed_steps": 17000,
-    "tune_objective": 0.4375542633059791
+    "total_token": 1974715,
+    "completed_steps": 1000,
+    "tune_objective": 0.39956710286966757
 }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0f294b9fb269ca49e980c3ffa101f250bc2cf3e76bad910a3fe9ca6640d8e6da
+oid sha256:6a667f9decf7af7cec90a666c53820a1019f056c5ce05dd606e613c926596526
 size 497774208

special_tokens_map.json

@@ -1,30 +1,6 @@
 {
-  "bos_token": {
-    "content": "<|endoftext|>",
-    "lstrip": false,
-    "normalized": true,
-    "rstrip": false,
-    "single_word": false
-  },
-  "eos_token": {
-    "content": "<|endoftext|>",
-    "lstrip": false,
-    "normalized": true,
-    "rstrip": false,
-    "single_word": false
-  },
-  "pad_token": {
-    "content": "<|endoftext|>",
-    "lstrip": false,
-    "normalized": true,
-    "rstrip": false,
-    "single_word": false
-  },
-  "unk_token": {
-    "content": "<|endoftext|>",
-    "lstrip": false,
-    "normalized": true,
-    "rstrip": false,
-    "single_word": false
-  }
+  "bos_token": "<|endoftext|>",
+  "eos_token": "<|endoftext|>",
+  "pad_token": "<|endoftext|>",
+  "unk_token": "<|endoftext|>"
 }

train_configs.json

@@ -29,7 +29,7 @@
     "seed": 0,
     "model_type": null,
     "block_size": 1024,
-    "mini_epoch": 1,
+    "mini_epoch": 3,
     "rollout_game": "baseline3v2",
     "rl_algo": "off_policy",
     "constraint_type": "kl",
@@ -39,7 +39,7 @@
     "n_generate": 4,
     "n_augment": 0,
     "gradient_checkpoint": false,
-    "group_relative_norm": true,
+    "group_relative_norm": false,
     "sample_config": {
         "do_sample": true,
         "min_new_tokens": 5,
@@ -52,13 +52,13 @@
         "max_new_tokens": 40
     },
     "rollout_config": {
-        "accuracy_w": 8764.345947696369,
+        "accuracy_w": 0.0005981501367285106,
         "len_pen": 1.0,
-        "accuracy_w2": 0.008731239835689814,
+        "accuracy_w2": 505.3115905942363,
         "len_pen2": 1.0,
-        "threshold": 0.00013728894342626908,
+        "threshold": 0.0001942205696972693,
         "similarity_fn": "rouge",
-        "acc_scale": "lin",
+        "acc_scale": "log",
         "max_ctx_len": 760
     },
     "ent_coef": 0.0001,
@@ -68,7 +68,7 @@
     "prompt_2": "Given the text: {role1_output}\nReconstruct the summarized text to the detailed:",
     "prompt_eval": "{text}\nTL;DR:",
     "epsilon": 0.2,
-    "a2g_norm": true,
+    "a2g_norm": false,
     "vllm_sleep": true,
     "lora": false,
     "need_attn_mask": true,
@@ -82,7 +82,7 @@
     "trunc_eval": 256,
     "buffer_max_size": 20000,
     "trunc_evals": [
-        "xsum___2000"
+        "xsum___20000"
     ],
     "use_deepspeed": false,
     "zero_config": 2,