add human-like QLoRA training config for Llama 3.1 8B

human_chat_qlora.yml

@@ -0,0 +1,86 @@
+# Llama 3.1 8B — Human-like QLoRA fine-tune
+#
+# Goal: natural, warm conversation; never corrects user errors; direct responses
+# Hardware: single RTX 5080 (16 GB VRAM)
+# Method: QLoRA (4-bit) via Axolotl
+#
+# Prerequisites:
+#   pip install -e '.[flash-attn]'   (inside your axolotl repo)
+#   huggingface-cli login            (meta-llama is a gated model)
+#
+# Run:
+#   axolotl train human_chat_qlora.yml
+#   axolotl merge-lora human_chat_qlora.yml   # (optional) merge adapter into base
+
+base_model: meta-llama/Meta-Llama-3.1-8B-Instruct
+model_type: LlamaForCausalLM
+tokenizer_type: AutoTokenizer
+
+load_in_4bit: true
+strict: false
+
+# --- System prompt baked into every conversation ---
+# This is the primary lever for "no error correcting, more human-like"
+chat_template: llama3
+default_system_message: >-
+  You are a direct, warm, and genuinely helpful assistant.
+  Respond to the user's intent naturally — never comment on typos, grammar,
+  or phrasing issues in their message. Just understand what they mean and give
+  a clear, useful, conversational answer as if talking to a knowledgeable friend.
+
+# --- Datasets ---
+# Capybara: ~16k carefully curated multi-turn conversations — best for natural tone
+# OpenHermes-2.5: broad instruction coverage — sampled to 5% to keep balance
+datasets:
+  - path: LDJnr/Capybara
+    type: sharegpt
+    conversation: llama3
+    split: train
+
+  - path: teknium/OpenHermes-2.5
+    type: sharegpt
+    conversation: llama3
+    split: "train[:5%]"
+
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
+output_dir: ./outputs/llama31-8b-humanchat
+
+sequence_len: 4096
+sample_packing: true
+pad_to_sequence_len: true
+
+# --- QLoRA adapter ---
+adapter: qlora
+lora_r: 64
+lora_alpha: 32
+lora_dropout: 0.05
+lora_target_linear: true
+
+# --- Training hyperparameters ---
+# Effective batch = micro_batch_size x gradient_accumulation = 2 x 4 = 8
+micro_batch_size: 2
+gradient_accumulation_steps: 4
+num_epochs: 2
+optimizer: paged_adamw_32bit
+lr_scheduler: cosine
+learning_rate: 2e-4
+warmup_ratio: 0.05
+weight_decay: 0.1
+
+train_on_inputs: false
+group_by_length: false
+bf16: auto
+tf32: false
+
+# --- Memory & speed ---
+gradient_checkpointing: true
+flash_attention: true
+
+# --- Logging & checkpointing ---
+logging_steps: 10
+evals_per_epoch: 2
+saves_per_epoch: 1
+
+special_tokens:
+  pad_token: "<|eot_id|>"