Upload checkpoints/step-00750 (omini_sauce_1)

checkpoints/step-00750/adapter_config.json

@@ -0,0 +1,50 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": null,
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "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": 64,
+  "lora_bias": false,
+  "lora_dropout": 0.0,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "monteclora_config": null,
+  "peft_type": "LORA",
+  "peft_version": "0.20.0",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "down",
+    "wv",
+    "wk",
+    "wo",
+    "up",
+    "gate",
+    "wq"
+  ],
+  "target_parameters": null,
+  "task_type": null,
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false,
+  "velora_config": null
+}

checkpoints/step-00750/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5d37177617884cdfb63b76805b2078f2b50744d85e7ec9e828680cbd834d77c
+size 457112440

checkpoints/step-00750/omini_sauce_1.toml

@@ -0,0 +1,55 @@
+# GROUNDEDSECRET — treino sério do krea2_omini_grounded (7000 steps).
+# Método validado no probe 250 + melhorias mapeadas em docs/OMINI_GROUNDED_SEGREDO.md:
+#   - caption_dropout 0.1 (treina o uncond grounded do CFG)
+#   - vl_grounding_jitter 384-768 (dial de inferência robusto)
+#   - multi-resolução 512/768/1024 (dataset é 1080p nativo)
+output_dir = '/workspace/checkpoints/omini_sauce_1'
+dataset = '/workspace/configs/omini_sauce_dataset.toml'
+epochs = 1000
+max_steps = 20000
+micro_batch_size_per_gpu = 1
+gradient_accumulation_steps = 1
+pipeline_stages = 1
+gradient_clipping = 1.0
+warmup_steps = 100
+save_every_n_steps = 250
+checkpoint_every_n_minutes = 60
+activation_checkpointing = true
+save_dtype = 'bfloat16'
+caching_batch_size = 1
+blocks_to_swap = 8
+
+[model]
+type = 'krea2_omini_grounded'
+diffusion_model = '/workspace/models/krea2/diffusion_models/krea2_raw_fp8_scaled.safetensors'
+vae = '/workspace/models/krea2/split_files/vae/qwen_image_vae.safetensors'
+text_encoders = [
+    {path = '/workspace/models/krea2/text_encoders/qwen3vl_4b_bf16.safetensors', type = 'krea2'},
+]
+dtype = 'bfloat16'
+diffusion_model_dtype = 'float8'
+timestep_sample_method = 'logit_normal'
+flux_shift = true
+
+[krea2_omini_grounded]
+position_mode = 'width_shift'
+reference_position_offset = 1.0
+reference_timestep = 'zero'
+condition_dropout = 0.0
+condition_only_lora = true
+vl_prompt_style = 'plain'
+vl_longest_side = 768
+vl_grounding_jitter = [384, 768]
+caption_dropout = 0.1
+
+[adapter]
+type = 'lora'
+rank = 64
+dtype = 'bfloat16'
+
+[optimizer]
+type = 'AdamW8bitKahan'
+lr = 1e-4
+betas = [0.9, 0.99]
+weight_decay = 0.01
+stabilize = false