Training in progress, step 100, checkpoint

checkpoint-100/README.md

@@ -1,6 +1,6 @@
 ---
 library_name: peft
-base_model: HuggingFaceH4/zephyr-7b-beta
+base_model: mistralai/Mixtral-8x7B-Instruct-v0.1
 ---
 
 # Model Card for Model ID

checkpoint-100/adapter_config.json

@@ -1,7 +1,7 @@
 {
   "alpha_pattern": {},
   "auto_mapping": null,
-  "base_model_name_or_path": "HuggingFaceH4/zephyr-7b-beta",
+  "base_model_name_or_path": "mistralai/Mixtral-8x7B-Instruct-v0.1",
   "bias": "none",
   "fan_in_fan_out": false,
   "inference_mode": true,
@@ -20,10 +20,10 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "q_proj",
-    "v_proj",
     "k_proj",
-    "o_proj"
+    "o_proj",
+    "v_proj",
+    "q_proj"
   ],
   "task_type": "CAUSAL_LM",
   "use_dora": false,

checkpoint-100/adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:96e8d01683a4f1dd43f963a096cd007c509e505bcd4c33d14a6bee6a397b17e4
+oid sha256:0a02f1947ab36f1fff09eb9a70d5c59e7422cfa2414db71c34b73281ebf1f79c
 size 54560368

checkpoint-100/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b66811faa2c5d0dac760eb6a8ae05ac3df86f401503ea5dcb5f2c1bdfb59ab9d
+oid sha256:f59d1470a8bb6b190be0ef0269a1eca8dac817719f679bee82a9cff3a2078f67
 size 109267450

checkpoint-100/scheduler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:845227c64f4739d4bbba122720ec7ac903273c61f209a8b9b524f4ffeef9433f
+oid sha256:b5c5d1c5dfacd2e118f646f270d4987c263d81d3756dd72c85b10cf85d21cad3
 size 1064

checkpoint-100/special_tokens_map.json

@@ -1,9 +1,4 @@
 {
-  "additional_special_tokens": [
-    "<unk>",
-    "<s>",
-    "</s>"
-  ],
   "bos_token": {
     "content": "<s>",
     "lstrip": false,

checkpoint-100/tokenizer_config.json

@@ -27,13 +27,9 @@
       "special": true
     }
   },
-  "additional_special_tokens": [
-    "<unk>",
-    "<s>",
-    "</s>"
-  ],
+  "additional_special_tokens": [],
   "bos_token": "<s>",
-  "chat_template": "{% for message in messages %}\n{% if message['role'] == 'user' %}\n{{ '<|user|>\n' + message['content'] + eos_token }}\n{% elif message['role'] == 'system' %}\n{{ '<|system|>\n' + message['content'] + eos_token }}\n{% elif message['role'] == 'assistant' %}\n{{ '<|assistant|>\n'  + message['content'] + eos_token }}\n{% endif %}\n{% if loop.last and add_generation_prompt %}\n{{ '<|assistant|>' }}\n{% endif %}\n{% endfor %}",
+  "chat_template": "{{ bos_token }}{% for message in messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% if message['role'] == 'user' %}{{ '[INST] ' + message['content'] + ' [/INST]' }}{% elif message['role'] == 'assistant' %}{{ message['content'] + eos_token}}{% else %}{{ raise_exception('Only user and assistant roles are supported!') }}{% endif %}{% endfor %}",
   "clean_up_tokenization_spaces": false,
   "eos_token": "</s>",
   "legacy": true,
@@ -44,7 +40,6 @@
   "sp_model_kwargs": {},
   "spaces_between_special_tokens": false,
   "tokenizer_class": "LlamaTokenizer",
-  "truncation_side": "left",
   "unk_token": "<unk>",
-  "use_default_system_prompt": true
+  "use_default_system_prompt": false
 }

checkpoint-100/trainer_state.json

@@ -1,6 +1,6 @@
 {
-  "best_metric": 0.022206587716937065,
-  "best_model_checkpoint": "./Zephyr/28-03-24-Weni-WeniGPT-QA-Zephyr-7B-5.0.0-KTO_WeniGPT Experiment using KTO trainer with no collator, Mixstral model and random system prompt.-2_max_steps-786_batch_32_2024-03-28_ppid_9/checkpoint-100",
+  "best_metric": 0.041092585772275925,
+  "best_model_checkpoint": "./mixstral/01-04-24-Weni-WeniGPT-QA-Mixstral-7B-5.0.0-KTO_WeniGPT Experiment using KTO trainer with no collator, Mixstral model and random system prompt.-2_max_steps-131_batch_32_2024-04-01_ppid_727/checkpoint-100",
   "epoch": 0.7604562737642585,
   "eval_steps": 50,
   "global_step": 100,
@@ -10,102 +10,102 @@
   "log_history": [
     {
       "epoch": 0.15,
-      "grad_norm": 3.2390341758728027,
-      "kl": 1.4365953207015991,
-      "learning_rate": 0.0001666666666666667,
-      "logps/chosen": NaN,
-      "logps/rejected": NaN,
-      "loss": 1.2323,
-      "rewards/chosen": NaN,
-      "rewards/margins": NaN,
-      "rewards/rejected": NaN,
-      "step": 20
+      "grad_norm": 1.3443933725357056,
+      "learning_rate": 0.00017480314960629923,
+      "loss": 0.3373,
+      "step": 20,
+      "train/kl": 0.27332180738449097,
+      "train/logps/chosen": -139.77467346191406,
+      "train/logps/rejected": -186.41754150390625,
+      "train/rewards/chosen": 0.8578631281852722,
+      "train/rewards/margins": 2.404827654361725,
+      "train/rewards/rejected": -1.5469645261764526
     },
     {
       "epoch": 0.3,
-      "grad_norm": 0.6226775050163269,
-      "kl": 0.1346512734889984,
-      "learning_rate": 0.00019580052493438322,
-      "logps/chosen": NaN,
-      "logps/rejected": NaN,
-      "loss": 0.1437,
-      "rewards/chosen": NaN,
-      "rewards/margins": NaN,
-      "rewards/rejected": NaN,
-      "step": 40
+      "grad_norm": 0.4922572672367096,
+      "learning_rate": 0.00014330708661417323,
+      "loss": 0.0944,
+      "step": 40,
+      "train/kl": 0.05194361135363579,
+      "train/logps/chosen": -141.4928741455078,
+      "train/logps/rejected": -236.62286376953125,
+      "train/rewards/chosen": 2.591707468032837,
+      "train/rewards/margins": 10.605849504470825,
+      "train/rewards/rejected": -8.014142036437988
     },
     {
       "epoch": 0.38,
-      "eval_kl": 0.0960722267627716,
-      "eval_logps/chosen": -158.278564453125,
-      "eval_logps/rejected": -400.1436767578125,
-      "eval_loss": 0.028233768418431282,
-      "eval_rewards/chosen": 5.2842488288879395,
-      "eval_rewards/margins": 25.394336700439453,
-      "eval_rewards/rejected": -20.11008644104004,
-      "eval_runtime": 214.1148,
-      "eval_samples_per_second": 2.232,
-      "eval_steps_per_second": 0.56,
+      "eval/kl": 0.05559438839554787,
+      "eval/logps/chosen": -123.70791625976562,
+      "eval/logps/rejected": -280.04022216796875,
+      "eval/rewards/chosen": 3.6042592525482178,
+      "eval/rewards/margins": 15.286608457565308,
+      "eval/rewards/rejected": -11.68234920501709,
+      "eval_loss": 0.05842626839876175,
+      "eval_runtime": 959.2742,
+      "eval_samples_per_second": 0.496,
+      "eval_steps_per_second": 0.124,
       "step": 50
     },
     {
       "epoch": 0.46,
-      "grad_norm": 1.9644193649291992,
-      "kl": 0.026479745283722878,
-      "learning_rate": 0.0001905511811023622,
-      "logps/chosen": NaN,
-      "logps/rejected": NaN,
-      "loss": 0.1405,
-      "rewards/chosen": NaN,
-      "rewards/margins": NaN,
-      "rewards/rejected": NaN,
-      "step": 60
+      "grad_norm": 1.1183429956436157,
+      "learning_rate": 0.00011181102362204725,
+      "loss": 0.0543,
+      "step": 60,
+      "train/kl": 0.0,
+      "train/logps/chosen": -123.6012191772461,
+      "train/logps/rejected": -282.85455322265625,
+      "train/rewards/chosen": 3.592074155807495,
+      "train/rewards/margins": 16.24874472618103,
+      "train/rewards/rejected": -12.656670570373535
     },
     {
       "epoch": 0.61,
-      "grad_norm": 0.28626948595046997,
-      "kl": 0.018679404631257057,
-      "learning_rate": 0.00018530183727034121,
-      "logps/chosen": NaN,
-      "logps/rejected": NaN,
-      "loss": 0.1041,
-      "rewards/chosen": NaN,
-      "rewards/margins": NaN,
-      "rewards/rejected": NaN,
-      "step": 80
+      "grad_norm": 0.41082897782325745,
+      "learning_rate": 8.031496062992126e-05,
+      "loss": 0.049,
+      "step": 80,
+      "train/kl": 0.0,
+      "train/logps/chosen": -113.79830932617188,
+      "train/logps/rejected": -275.3311767578125,
+      "train/rewards/chosen": 3.7077345848083496,
+      "train/rewards/margins": 15.970922946929932,
+      "train/rewards/rejected": -12.263188362121582
     },
     {
       "epoch": 0.76,
-      "grad_norm": 0.0979766920208931,
-      "kl": 0.0,
-      "learning_rate": 0.00018005249343832022,
-      "logps/chosen": NaN,
-      "logps/rejected": NaN,
-      "loss": 0.0615,
-      "rewards/chosen": NaN,
-      "rewards/margins": NaN,
-      "rewards/rejected": NaN,
-      "step": 100
+      "grad_norm": 0.810788094997406,
+      "learning_rate": 4.881889763779528e-05,
+      "loss": 0.0446,
+      "step": 100,
+      "train/kl": 0.02022113837301731,
+      "train/logps/chosen": -120.12004852294922,
+      "train/logps/rejected": -306.6509704589844,
+      "train/rewards/chosen": 3.6804699897766113,
+      "train/rewards/margins": 18.751341342926025,
+      "train/rewards/rejected": -15.070871353149414
     },
     {
       "epoch": 0.76,
-      "eval_kl": 0.5886130928993225,
-      "eval_logps/chosen": -153.61863708496094,
-      "eval_logps/rejected": -383.54296875,
-      "eval_loss": 0.022206587716937065,
-      "eval_rewards/chosen": 5.750241756439209,
-      "eval_rewards/margins": 24.20025062561035,
-      "eval_rewards/rejected": -18.450010299682617,
-      "eval_runtime": 213.9917,
-      "eval_samples_per_second": 2.234,
-      "eval_steps_per_second": 0.561,
+      "eval/kl": 0.0,
+      "eval/logps/chosen": -122.67949676513672,
+      "eval/logps/rejected": -363.1288146972656,
+      "eval/rewards/chosen": 3.7070999145507812,
+      "eval/rewards/margins": 23.698307037353516,
+      "eval/rewards/rejected": -19.991207122802734,
+      "eval_loss": 0.041092585772275925,
+      "eval_runtime": 961.5768,
+      "eval_samples_per_second": 0.495,
+      "eval_steps_per_second": 0.124,
       "step": 100
     }
   ],
   "logging_steps": 20,
-  "max_steps": 786,
+  "max_steps": 131,
   "num_input_tokens_seen": 0,
-  "num_train_epochs": 6,
+  "num_train_epochs": 1,
   "save_steps": 100,
   "total_flos": 0.0,
   "train_batch_size": 4,

checkpoint-100/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2143b63cfb768928620763e12d44a1780f6206ec9bca22cf2aab4ebe0626cbeb
+oid sha256:9c3a1382759b0a828826f90087526cfc46185b8f0940950b7995edb66ba51fff
 size 5752