Upload folder using huggingface_hub

step5500/adapter_config.json

@@ -0,0 +1,47 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": null,
+  "bias": "none",
+  "corda_config": null,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": false,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "to_k",
+    "to_v",
+    "to_add_out",
+    "add_k_proj",
+    "to_out.0",
+    "add_v_proj",
+    "proj",
+    "net.2",
+    "txt_mod.1",
+    "add_q_proj",
+    "img_mod.1",
+    "to_q"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

step5500/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1508bc39cecdd9aa5816e4c4d21a143e2c2c1ece330749eafc9f92b81eefd8e8
+size 2359534496

step5500/qwen_lora_config.toml

@@ -0,0 +1,52 @@
+# Main training settings
+output_dir = '/workspace/qwen-lora-output'
+dataset = '/workspace/qwen_dataset.toml'
+epochs = 10000
+max_steps = 6000
+micro_batch_size_per_gpu = 1
+pipeline_stages = 1
+gradient_accumulation_steps = 1
+gradient_clipping = 1.0
+warmup_steps = 200
+save_every_n_steps = 250
+checkpoint_every_n_minutes = 120
+activation_checkpointing = true
+
+eval_every_n_steps = 500
+eval_before_first_step = true
+eval_micro_batch_size_per_gpu = 1
+eval_gradient_accumulation_steps = 1
+
+
+[model]
+type = 'qwen_image'
+# This path is used to find the tokenizer config files.
+diffusers_path = '/workspace/models/Qwen-Image'
+
+# Explicitly define paths to each core model component using the correct key names.
+transformer_path = '/workspace/models/qwen_transformer/split_files/diffusion_models/qwen_image_bf16.safetensors'
+text_encoder_path = '/workspace/models/qwen_text_encoder/split_files/text_encoders/qwen_2.5_vl_7b.safetensors'
+vae_path = '/workspace/models/Qwen-Image/vae/diffusion_pytorch_model.safetensors'
+
+# BF16 is great for L40S, and fp8 for the transformer saves VRAM with minimal quality loss.
+dtype = 'bfloat16'
+transformer_dtype = 'bfloat16'
+timestep_sample_method = 'logit_normal'
+
+[adapter]
+type = 'lora'
+rank = 128
+dtype = 'bfloat16'
+
+[optimizer]
+type = 'adamw_optimi'
+lr = 5e-5
+betas = [0.9, 0.99]
+weight_decay = 0.01
+eps = 1e-8
+
+[monitoring]
+enable_wandb = true
+wandb_api_key = 'cc3f7d9c61eecb958c5b52f48087ecef56cc32b4' # Paste your key here
+wandb_tracker_name = 'Qwen-Image-Dlay_LoRA'
+wandb_run_name = 'L40S-Run-1'