- Add KTO Post-training example

post-training/README.md

@@ -40,6 +40,14 @@ python tools/datasets/preprocess_data_with_chat_template.py --input data/sft/lla
 python tools/datasets/preprocess_data_with_chat_template.py --input data/sft/llama3_sft_train_filtered.jsonl --output-prefix data/sft/llama3_val --tokenizer-path checkpoints/neox_converted/llama3-8b-instruct/tokenizer --jsonl-keys messages
 ```
 
+## KTO data
+```bash
+python post-training/llama_dpo_data.py
+python tools/datasets/preprocess_data_with_chat_template.py --input data/kto/llama3_sft_train_filtered.jsonl --output-prefix data/kto/llama3_train --tokenizer-path checkpoints/neox_converted/llama3-8b-instruct/tokenizer --jsonl-keys messages --reward-key reward
+python tools/datasets/preprocess_data_with_chat_template.py --input data/kto/llama3_sft_test_filtered.jsonl --output-prefix data/kto/llama3_test --tokenizer-path checkpoints/neox_converted/llama3-8b-instruct/tokenizer --jsonl-keys messages --reward-key reward
+python tools/datasets/preprocess_data_with_chat_template.py --input data/kto/llama3_sft_train_filtered.jsonl --output-prefix data/kto/llama3_val --tokenizer-path checkpoints/neox_converted/llama3-8b-instruct/tokenizer --jsonl-keys messages --reward-key reward
+```
+
 
 ## Converting back to hf
 ```bash

post-training/configs/llama3-8b-kto.yml

@@ -0,0 +1,120 @@
+{
+  "pipe_parallel_size": 0,
+  "model_parallel_size": 4,
+  "make_vocab_size_divisible_by": 1,
+
+  # model settings
+  "num_layers": 32,
+  "hidden_size": 4096,
+  "num_attention_heads": 32,
+  "num_kv_heads": 8,
+  # llama3 supports more than this but this is just for testing.
+  "seq_length": 1024,
+  "max_position_embeddings": 1024,
+  "pos_emb": "rotary",
+  "rotary_pct": 1,
+  "rotary_emb_base": 500000,
+  "rope_fusion": true,
+  "no_weight_tying": true,
+  "gpt_j_residual": false,
+  "output_layer_parallelism": "column",
+  "norm": "rmsnorm",
+  "rms_norm_epsilon": 1.0e-5,
+
+  "attention_config": [[["flash"], 32]],
+
+  "scaled_upper_triang_masked_softmax_fusion": true,
+  "bias_gelu_fusion": false,
+  "use_bias_in_norms": false,
+  "use_bias_in_attn_linear": false,
+  "use_bias_in_mlp": false,
+  "use_flashattn_swiglu": true,
+  "activation": "swiglu",
+  "intermediate_size": 14336,
+  "mlp_multiple_of": 14336,
+
+   "optimizer": {
+     "type": "Adam",
+     "params": {
+       "lr": 0.00001,
+       "betas": [0.9, 0.95],
+       "eps": 1.0e-8
+     }
+   },
+  "min_lr": 0.000001,
+
+   "zero_optimization": {
+    "stage": 1,
+    "allgather_partitions": true,
+    "allgather_bucket_size": 1260000000,
+    "overlap_comm": true,
+    "reduce_scatter": true,
+    "reduce_bucket_size": 1260000000,
+    "contiguous_gradients": true,
+    "cpu_offload": false
+  },
+
+
+  "train_impl": "kto",
+  "kto_fp32": true,
+  "kto_beta": 0.1,
+  "allow_chopped": false,
+  "train_label_data_paths": [ "data/kto/llama3_train_messages_label_document" ],
+  "test_label_data_paths": [ "data/kto/llama3_test_messages_label_document" ],
+  "valid_label_data_paths": [ "data/kto/llama3_train_messages_label_document" ],
+  "train_data_paths": [ "data/kto/llama3_train_messages_document" ],
+  "test_data_paths": [ "data/kto/llama3_test_messages_document" ],
+  "valid_data_paths": [ "data/kto/llama3_train_messages_document" ],
+  "train_reward_data_paths": [ "data/kto/llama3_train_messages_reward_document" ],
+  "test_reward_data_paths": [ "data/kto/llama3_test_messages_reward_document" ],
+  "valid_reward_data_paths": [ "data/kto/llama3_train_messages_reward_document" ],
+
+  "train_micro_batch_size_per_gpu": 32,
+  "gradient_accumulation_steps": 2,
+  "data_impl": "mmap",
+  "pack_impl": "unpacked",
+  "num_workers": 1,
+
+  "checkpoint_activations": true,
+  "checkpoint_num_layers": 1,
+  "partition_activations": true,
+  "synchronize_each_layer": true,
+
+  "gradient_clipping": 1.0,
+  "weight_decay": 0.1,
+  "hidden_dropout": 0,
+  "attention_dropout": 0,
+
+  "precision": "bfloat16",
+  "fp32_allreduce": true,
+  "bf16": {
+    "enabled": true
+  },
+  "data_types": {
+    "grad_accum_dtype": "fp32"
+  },
+
+  "train_iters": 477,
+  "lr_decay_iters": 477,
+  "distributed_backend": "nccl",
+  "lr_decay_style": "cosine",
+  "warmup": 0.1,
+  "checkpoint_factor": 1000,
+  "eval_interval": 100,
+  "eval_iters": 10,
+
+  "log_interval": 1,
+  "steps_per_print": 1,
+  "wall_clock_breakdown": true,
+
+
+  "save": "checkpoints/kto/llama3/llama3-8b-instruct",
+  #"load": "", # once run is started, to restart from intermediate ckpt use "load" = "save"
+  "load": "checkpoints/neox_converted/llama3-8b-instruct",
+  "vocab-file": "checkpoints/neox_converted/llama3-8b-instruct/tokenizer/tokenizer.json",
+  "use_wandb": true,
+  "wandb_group": "llama3-8b-instruct",
+  "wandb_project": "ultrafeedback-kto",
+  "finetune": true, # set to false once resuming from intermediate finetuning step
+  "tokenizer_type": "HFTokenizer",
+}

post-training/llama_data.py

@@ -22,7 +22,6 @@
     ) as f:
         writer = jsonlines.Writer(f)
         for item in raw_datasets[split]:
-            # add empty system messages
             item["chosen"] = item["chosen"]
             item["rejected"] = item["rejected"]
             writer.write(item)
@@ -33,6 +32,18 @@
     ) as f:
         writer = jsonlines.Writer(f)
         for item in raw_datasets[split]:
-            # add empty system messages
             item["messages"] = item["chosen"]
             writer.write(item)
+os.makedirs(os.path.join("data", "kto"), exist_ok=True)
+for split in ["train", "test"]:
+    with open(
+        os.path.join("data", "kto", f"llama3_kto_{split}_filtered.jsonl"), "w"
+    ) as f:
+        writer = jsonlines.Writer(f)
+        for item in raw_datasets[split]:
+            item["messages"] = item["chosen"]
+            item["reward"] = 1
+            writer.write(item)
+            item["messages"] = item["rejected"]
+            item["reward"] = -1
+            writer.write(item)