upload

README.md

@@ -3,4 +3,8 @@ license: mit
 library_name: transformers
 tags:
 - autoencoder
----
+---
+
+This model is converted from the pretrained VQGAN of Meta's [Chameleon](https://huggingface.co/facebook/chameleon-7b). We adapt the VQGAN model to our [Being-VL](https://github.com/BeingBeyond/Being-VL-0.5) architecture. The license for this VQGAN model follows the original license of [Chameleon](https://ai.meta.com/resources/models-and-libraries/chameleon-license).
+
+To use this model, please refer to the [Being-VL documentation](https://github.com/BeingBeyond/Being-VL-0.5).

config.json

@@ -0,0 +1,40 @@
+{
+  "architectures": [
+    "ChameleonVQVAE"
+  ],
+  "attn_resolutions": [],
+  "attn_type": "vanilla",
+  "base_channels": 128,
+  "ch": 128,
+  "ch_mult": [
+    1,
+    1,
+    2,
+    2,
+    4
+  ],
+  "channel_multiplier": [
+    1,
+    1,
+    2,
+    2,
+    4
+  ],
+  "double_latent": false,
+  "double_z": false,
+  "dropout": 0.0,
+  "embed_dim": 256,
+  "in_channels": 3,
+  "initializer_range": 0.02,
+  "latent_channels": 256,
+  "model_type": "chameleon_vqgan",
+  "n_embed": 8192,
+  "num_embeddings": 8192,
+  "num_res_blocks": 2,
+  "out_ch": 3,
+  "out_channels": 3,
+  "resolution": 512,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.45.0.dev0",
+  "z_channels": 256
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c2b651489d0a1076d51cb81b7afcf107d8352b555119246e14473470fffedbd9
+size 137467246