Upload fully merged KielGen-Pro model (16-bit safetensors & 4-bit ready)

README.md

@@ -1,13 +1,51 @@
 ---
 base_model: deepseek-ai/Janus-Pro-1B
-library_name: peft
+library_name: transformers
 tags:
   - multimodal
-  - lora
-  - kielgen-pro
   - janus-pro
+  - kielgen-pro
+  - lora-merged
+  - 4-bit
+  - 16-bit
 ---
 
-# KielGen-Pro Phase 1 LoRA Adapters
+# KielGen-Pro (Merged 16-bit & 4-bit Ready)
+
+This repository contains the fully merged **KielGen-Pro** multimodal model, fine-tuned from `Janus-Pro-1B` on a custom 10,000-sample curriculum.
+
+## Deployment Instructions
+
+### Load in 16-Bit (Full Precision)
+```python
+import torch
+from janus.models import MultiModalityCausalLM
+
+model = MultiModalityCausalLM.from_pretrained(
+    "kiel2/KielGen-Pro",
+    torch_dtype=torch.float16,
+    device_map="auto",
+    trust_remote_code=True
+)
+```
+
+### Load in Memory-Efficient 4-Bit
+```python
+import torch
+from janus.models import MultiModalityCausalLM
+from transformers import BitsAndBytesConfig
+
+quantization_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_compute_dtype=torch.float16,
+    bnb_4bit_quant_type="nf4",
+    bnb_4bit_use_double_quant=True
+)
 
-This repository contains Phase 1 LoRA weights for **KielGen-Pro**, fine-tuned on a multi-domain curriculum (UI/Code, Artistic styling, and Reasoning) using `Janus-Pro-1B`.
+model = MultiModalityCausalLM.from_pretrained(
+    "kiel2/KielGen-Pro",
+    quantization_config=quantization_config,
+    device_map="auto",
+    trust_remote_code=True
+)
+```

config.json

@@ -0,0 +1,83 @@
+{
+  "aligner_config": {
+    "cls": "MlpProjector",
+    "model_type": "aligner",
+    "params": {
+      "depth": 2,
+      "input_dim": 1024,
+      "n_embed": 2048,
+      "projector_type": "mlp_gelu"
+    }
+  },
+  "architectures": [
+    "MultiModalityCausalLM"
+  ],
+  "dtype": "float16",
+  "gen_aligner_config": {
+    "cls": "MlpProjector",
+    "model_type": "gen_aligner",
+    "params": {
+      "depth": 2,
+      "input_dim": 8,
+      "n_embed": 2048,
+      "projector_type": "mlp_gelu"
+    }
+  },
+  "gen_head_config": {
+    "cls": "vision_head",
+    "model_type": "gen_head",
+    "params": {
+      "image_token_embed": 2048,
+      "image_token_size": 16384,
+      "n_embed": 2048
+    }
+  },
+  "gen_vision_config": {
+    "cls": "VQ-16",
+    "model_type": "gen_vision",
+    "params": {
+      "image_token_size": 16384,
+      "n_embed": 8
+    }
+  },
+  "language_config": {
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "bos_token_id": 1,
+    "dtype": "bfloat16",
+    "eos_token_id": 2,
+    "head_dim": 128,
+    "hidden_act": "silu",
+    "hidden_size": 2048,
+    "initializer_range": 0.02,
+    "intermediate_size": 5632,
+    "max_position_embeddings": 16384,
+    "mlp_bias": false,
+    "model_type": "llama",
+    "num_attention_heads": 16,
+    "num_hidden_layers": 24,
+    "num_key_value_heads": 16,
+    "pad_token_id": null,
+    "pretraining_tp": 1,
+    "rms_norm_eps": 1e-06,
+    "rope_parameters": {
+      "rope_theta": 10000.0,
+      "rope_type": "default"
+    },
+    "tie_word_embeddings": false,
+    "use_cache": true,
+    "vocab_size": 102400
+  },
+  "model_type": "multi_modality",
+  "transformers_version": "5.16.1",
+  "vision_config": {
+    "cls": "CLIPVisionTower",
+    "model_type": "vision",
+    "params": {
+      "image_size": 384,
+      "model_name": "siglip_large_patch16_384",
+      "select_feature": "same",
+      "select_layer": -1
+    }
+  }
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ae45710ef842ae7428731192c407fd9d5b45c967f40804230b0f4f2736ec732
+size 4178705502

processor_config.json

@@ -0,0 +1,34 @@
+{
+  "add_special_token": false,
+  "ignore_id": -100,
+  "image_end_tag": "<end_of_image>",
+  "image_processor": {
+    "background_color": [
+      127,
+      127,
+      127
+    ],
+    "do_normalize": true,
+    "image_mean": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "image_processor_type": "VLMImageProcessor",
+    "image_size": 384,
+    "image_std": [
+      0.5,
+      0.5,
+      0.5
+    ],
+    "min_size": 14,
+    "rescale_factor": 0.00392156862745098
+  },
+  "image_start_tag": "<begin_of_image>",
+  "image_tag": "<image_placeholder>",
+  "mask_prompt": true,
+  "num_image_tokens": 576,
+  "pad_tag": "<\uff5c\u2581pad\u2581\uff5c>",
+  "processor_class": "VLChatProcessor",
+  "sft_format": "deepseek"
+}

tokenizer_config.json

@@ -0,0 +1,26 @@
+{
+  "add_prefix_space": null,
+  "backend": "tokenizers",
+  "bos_token": "<｜begin▁of▁sentence｜>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<｜end▁of▁sentence｜>",
+  "extra_special_tokens": [
+    "<image_placeholder>",
+    "<patch_placeholder>",
+    "<|ref|>",
+    "<|/ref|>",
+    "<|det|>",
+    "<|/det|>",
+    "<|grounding|>",
+    "<|User|>",
+    "<|Assistant|>"
+  ],
+  "is_local": false,
+  "local_files_only": false,
+  "model_max_length": 16384,
+  "pad_token": "<｜▁pad▁｜>",
+  "processor_class": "VLChatProcessor",
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": null,
+  "use_default_system_prompt": true
+}