Upload HfMoondream

README.md

@@ -7,7 +7,7 @@ Moondream is a small vision language model designed to run efficiently everywher
 
 [Website](https://moondream.ai/) / [Demo](https://moondream.ai/playground) / [GitHub](https://github.com/vikhyat/moondream)
 
-This repository contains the 2025-04-14 **4bit** release of Moondream. On an Nvidia RTX 3090, it uses 2,305 MB of VRAM and runs at a speed of 187 tokens/second. We used quantization-aware training techniques to build this version of the model, allowing us to achieve a 45% reduction in memory usage with only an (TK)% drop in accuracy.
+This repository contains the 2025-04-14 **4bit** release of Moondream. On an Nvidia RTX 3090, it uses 2,305 MB of VRAM and runs at a speed of 187 tokens/second. We used quantization-aware training techniques to build this version of the model, allowing us to achieve a 45% reduction in memory usage with only an 2% drop in accuracy.
 
 There's more information about this version of the model in our [release blog post](https://moondream.ai/blog/smaller-faster-moondream-with-qat). Other revisions, as well as release history, can be found [here](https://huggingface.co/vikhyatk/moondream2).
 

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:325876fadb939f7c65f545d5d37b03f5035681b87bad1073f6d2e804ce2f4068
-size 1881750512
+oid sha256:dfce186edf359fff98d0c077ae389b980b6cae99279d157fc00b2d03ca65968f
+size 2032380848

text.py

@@ -187,7 +187,7 @@ def build_text_model(config: TextConfig, dtype: torch.dtype) -> nn.Module:
                 ]
             ),
             "post_ln": nn.LayerNorm(config.dim, dtype=dtype),
-            "lm_head": linear_cls(config.dim, config.vocab_size, dtype=dtype),
+            "lm_head": nn.Linear(config.dim, config.vocab_size, dtype=dtype),
         }
     )
     text.wte = nn.Parameter(torch.empty(config.vocab_size, config.dim, dtype=dtype))