Update model, now trined with OpenAssistant dataset in ChatML format

README.md

@@ -5,45 +5,66 @@ tags:
   - text-generation
 base_model: Locutusque/TinyMistral-248M
 datasets:
-  - WizardLM/WizardLM_evol_instruct_V2_196k
-  - KnutJaegersberg/WizardLM_evol_instruct_V2_196k_instruct_format
+  - OpenAssistant/oasst_top1_2023-08-25
 widget:
   - text: |-
-      ### Instruction:
-      Write the specs of a game about trolls and warriors in a fantasy world.
-      
-      ### Response:
-      The game is an adventure game that takes place on a planet, where players must explore their own unique abilities to survive. Players can use different strategies such as collecting items or trading them for gold or silver coins, but they also need to learn how to deal with obstacles and find new ways to escape.
-      
-      ### Instruction:
-      Tell me something curious about the Earth.
-      
-      ### Response:
-      The planet is a large, rocky world with an atmosphere of 10 billion years old and a surface area around 25 million miles (36 million kilometers) wide.
-      
-      ### Instruction:
-      What are some potential applications for quantum computing?
-      
-      ### Response:
+      <|im_start|>user
+      Write the specs of a game about trolls and warriors in a fantasy world.<|im_end|>
+      <|im_start|>assistant
+      The game is an adventure game that takes place on a planet, where players must explore their unique abilities to survive. Players can use different strategies such as collecting items or trading them for gold or silver coins, but they also need to learn how to deal with obstacles and find new ways to escape.<|im_end|>
+      <|im_start|>user
+      Could you tell me something curious about the Earth?<|im_end|>
+      <|im_start|>assistant
+      The planet is a large, rocky world with an atmosphere of 10 billion years old and a surface area around 25 million miles (36 million kilometers) wide.<|im_end|>
+      <|im_start|>user
+      What are some potential applications for quantum computing?<|im_end|>
+      <|im_start|>assistant
 inference:
   parameters:
     max_new_tokens: 64
     repetition_penalty: 1.18
 ---
 
-# Locutusque's TinyMistral-248M trained on the Evol Instruct dataset
+# Locutusque's TinyMistral-248M trained on OpenAssistant TOP-1 Conversation Threads
 
-- Base model: [Locutusque/TinyMistral-248M](https://huggingface.co/Locutusque/TinyMistral-248M)
-- Dataset: [WizardLM/WizardLM_evol_instruct_V2_196k](https://huggingface.co/datasets/WizardLM/WizardLM_evol_instruct_V2_196k)
-- Trained with [AutoTrain Advanced](https://github.com/huggingface/autotrain-advanced) using [these parameters](https://huggingface.co/Felladrin/TinyMistral-248M-Evol-Instruct/blob/321787e81e2eb0392d7ce2715154fb9c254e39b1/training_params.json) and [this CSV file](https://huggingface.co/datasets/KnutJaegersberg/WizardLM_evol_instruct_V2_196k_instruct_format/blob/93aa373501f829449f23efc91b3ac6e7a60a4d70/all_instructions.csv)
-- Availability in other ML formats:
-  - GGUF: [Felladrin/gguf-TinyMistral-248M-Evol-Instruct](https://huggingface.co/Felladrin/gguf-TinyMistral-248M-Evol-Instruct)
+- Base model: [Locutusque/TinyMistral-248M](https://huggingface.co/Locutusque/TinyMistral-248M/blob/90b89d18fdf27937dc04ab8a9b543c5af2991c7f/README.md)
+- Dataset: [OpenAssistant/oasst_top1_2023-08-25](https://huggingface.co/datasets/OpenAssistant/oasst_top1_2023-08-25)
 
 ## Recommended Prompt Format
 
 ```
-### Instruction:
-<instruction>
+<|im_start|>user
+{message}<|im_end|>
+<|im_start|>assistant
+```
+
+## How it was trained
+
+```ipython
+%pip install autotrain-advanced
+
+!autotrain setup
 
-### Response:
+!autotrain llm \
+    --train \
+    --trainer "sft" \
+    --model './TinyMistral-248M/' \
+    --model_max_length 4096 \
+    --block-size 1024 \
+    --project-name 'trained-model' \
+    --data-path "OpenAssistant/oasst_top1_2023-08-25" \
+    --train_split "train" \
+    --valid_split "test" \
+    --text-column "text" \
+    --lr 1e-5 \
+    --train_batch_size 2 \
+    --epochs 5 \
+    --evaluation_strategy "steps" \
+    --save-strategy "steps" \
+    --save-total-limit 2 \
+    --warmup-ratio 0.05 \
+    --weight-decay 0.0 \
+    --gradient-accumulation 8 \
+    --logging-steps 10 \
+    --scheduler "constant"
 ```

config.json

@@ -1,11 +1,10 @@
 {
-  "_name_or_path": "Locutusque/TinyMistral-248M",
+  "_name_or_path": "./TinyMistral-248M/",
   "architectures": [
     "MistralForCausalLM"
   ],
-  "bos_token_id": 32000,
-  "eos_token_id": 32001,
-  "pad_token_id": 32002,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
   "hidden_act": "silu",
   "hidden_size": 1024,
   "initializer_range": 0.02,
@@ -19,8 +18,8 @@
   "rope_theta": 10000.0,
   "sliding_window": 32,
   "tie_word_embeddings": false,
-  "torch_dtype": "float16",
+  "torch_dtype": "float32",
   "transformers_version": "4.34.1",
-  "use_cache": true,
+  "use_cache": false,
   "vocab_size": 32003
 }

generation_config.json

@@ -1,7 +1,6 @@
 {
   "_from_model_config": true,
-  "bos_token_id": 32000,
-  "eos_token_id": 32001,
-  "pad_token_id": 32002,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
   "transformers_version": "4.34.1"
 }

model.safetensors → pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5b06f829f8156b3724f0f7d8f10c75a09c08c9dc3655efa72272991dfe521597
-size 496052392
+oid sha256:1e01fe28eb61f7d317f3236e36c88be43e136ef4ae66a424e8c912a98141aba2
+size 992115782

special_tokens_map.json

@@ -1,13 +1,13 @@
 {
   "bos_token": {
-    "content": "<|bos|>",
+    "content": "<s>",
     "lstrip": false,
     "normalized": false,
     "rstrip": false,
     "single_word": false
   },
   "eos_token": {
-    "content": "<|endoftext|>",
+    "content": "</s>",
     "lstrip": false,
     "normalized": false,
     "rstrip": false,

tokenizer.json

@@ -80,7 +80,7 @@
     "single": [
       {
         "SpecialToken": {
-          "id": "<|bos|>",
+          "id": "<s>",
           "type_id": 0
         }
       },
@@ -94,7 +94,7 @@
     "pair": [
       {
         "SpecialToken": {
-          "id": "<|bos|>",
+          "id": "<s>",
           "type_id": 0
         }
       },
@@ -106,7 +106,7 @@
       },
       {
         "SpecialToken": {
-          "id": "<|bos|>",
+          "id": "<s>",
           "type_id": 1
         }
       },
@@ -118,13 +118,13 @@
       }
     ],
     "special_tokens": {
-      "<|bos|>": {
-        "id": "<|bos|>",
+      "<s>": {
+        "id": "<s>",
         "ids": [
-          32000
+          1
         ],
         "tokens": [
-          "<|bos|>"
+          "<s>"
         ]
       }
     }

tokenizer_config.json

@@ -1,6 +1,4 @@
 {
-  "add_bos_token": true,
-  "add_eos_token": false,
   "added_tokens_decoder": {
     "0": {
       "content": "<unk>",
@@ -52,15 +50,19 @@
     }
   },
   "additional_special_tokens": [],
-  "bos_token": "<|bos|>",
+  "bos_token": "<s>",
   "clean_up_tokenization_spaces": false,
-  "eos_token": "<|endoftext|>",
+  "eos_token": "</s>",
   "legacy": true,
-  "model_max_length": 1000000000000000019884624838656,
+  "max_length": 1536,
+  "model_max_length": 4096,
   "pad_token": "[PAD]",
   "sp_model_kwargs": {},
   "spaces_between_special_tokens": false,
+  "stride": 0,
   "tokenizer_class": "LlamaTokenizer",
+  "truncation_side": "right",
+  "truncation_strategy": "longest_first",
   "unk_token": "<unk>",
   "use_default_system_prompt": true
 }

training_params.json

@@ -1,47 +0,0 @@
-{
-    "model": "Felladrin/TinyMistral-248M-Evol-Instruct",
-    "data_path": "data/",
-    "project_name": "TinyMistral-248M-Evol-Instruct",
-    "train_split": "train",
-    "valid_split": null,
-    "text_column": "text",
-    "rejected_text_column": "rejected",
-    "token": null,
-    "lr": 0.0002,
-    "epochs": 1,
-    "batch_size": 12,
-    "warmup_ratio": 0.1,
-    "gradient_accumulation": 4,
-    "optimizer": "adamw_torch",
-    "scheduler": "linear",
-    "weight_decay": 0.01,
-    "max_grad_norm": 1.0,
-    "seed": 42,
-    "add_eos_token": false,
-    "block_size": 1024,
-    "use_peft": false,
-    "lora_r": 16,
-    "lora_alpha": 32,
-    "lora_dropout": 0.05,
-    "logging_steps": -1,
-    "evaluation_strategy": "epoch",
-    "save_total_limit": 1,
-    "save_strategy": "epoch",
-    "auto_find_batch_size": false,
-    "fp16": false,
-    "push_to_hub": false,
-    "use_int8": true,
-    "model_max_length": 1024,
-    "repo_id": null,
-    "use_int4": false,
-    "trainer": "sft",
-    "target_modules": null,
-    "merge_adapter": true,
-    "username": null,
-    "use_flash_attention_2": false,
-    "log": "none",
-    "disable_gradient_checkpointing": false,
-    "model_ref": null,
-    "dpo_beta": 0.1,
-    "prompt_text_column": "prompt"
-}