Merge 1a0ba385eb into 7050ed204b

The `cuda12-pip` package was wrongly named `cuda12_pip`
in requirements.txt

.gitignore

@@ -0,0 +1,2 @@
+checkpoints/*
+!checkpoints/README.md

README.md

@@ -2,7 +2,8 @@
 
 This repository contains JAX example code for loading and running the Grok-1 open-weights model.
 
-Make sure to download the checkpoint and place `ckpt-0` directory in `checkpoint`.
+Make sure to download the checkpoint and place the `ckpt-0` directory in `checkpoints` - see [Downloading the weights](#downloading-the-weights)
+
 Then, run
 
 ```shell
@@ -17,13 +18,37 @@ The script loads the checkpoint and samples from the model on a test input.
 Due to the large size of the model (314B parameters), a machine with enough GPU memory is required to test the model with the example code.
 The implementation of the MoE layer in this repository is not efficient. The implementation was chosen to avoid the need for custom kernels to validate the correctness of the model.
 
+# Model Specifications
+
+Grok-1 is currently designed with the following specifications:
+
+- **Parameters:** 314B
+- **Architecture:** Mixture of 8 Experts (MoE)
+- **Experts Utilization:** 2 experts used per token
+- **Layers:** 64
+- **Attention Heads:** 48 for queries, 8 for keys/values
+- **Embedding Size:** 6,144
+- **Tokenization:** SentencePiece tokenizer with 131,072 tokens
+- **Additional Features:**
+  - Rotary embeddings (RoPE)
+  - Supports activation sharding and 8-bit quantization
+- **Maximum Sequence Length (context):** 8,192 tokens
+
 # Downloading the weights
 
 You can download the weights using a torrent client and this magnet link:
+
 ```
 magnet:?xt=urn:btih:5f96d43576e3d386c9ba65b883210a393b68210e&tr=https%3A%2F%2Facademictorrents.com%2Fannounce.php&tr=udp%3A%2F%2Ftracker.coppersurfer.tk%3A6969&tr=udp%3A%2F%2Ftracker.opentrackr.org%3A1337%2Fannounce
 ```
 
+or directly using [HuggingFace 🤗 Hub](https://huggingface.co/xai-org/grok-1):
+```
+git clone https://github.com/xai-org/grok-1.git && cd grok-1
+pip install huggingface_hub[hf_transfer]
+huggingface-cli download xai-org/grok-1 --repo-type model --include ckpt-0/* --local-dir checkpoints --local-dir-use-symlinks False
+```
+
 # License
 
 The code and associated Grok-1 weights in this release are licensed under the

model.py

@@ -1002,7 +1002,7 @@ class DenseBlock(hk.Module):
             sharding=P("model", "data"),
             mesh=self.mesh,
             shard_axis=1,
-        )(h_w1 * h_v)
+        )(h_w1 * h_v) # TODO: Document why this isn't sequential and whether it should be.
 
         return h_dense
 
@@ -1036,13 +1036,10 @@ class DecoderLayer(hk.Module):
     ) -> DecoderOutput:
         """Transforms input embedding sequences to output embedding sequences."""
 
-        def layer_norm(x):
-            return hk_rms_norm(x)
-
+        sharding = P(self.data_axis, None)
         if self.shard_activations:
             sharding = P(self.data_axis, None, self.model_axis)
-        else:
-            sharding = P(self.data_axis, None)
+            
         h = with_sharding_constraint(inputs, sharding)
 
         attn_output = MHABlock(
@@ -1054,8 +1051,8 @@ class DecoderLayer(hk.Module):
             data_axis=self.data_axis,
             model_axis=self.model_axis,
         )(layer_norm(h), mask, layer_memory)
-        h_attn = attn_output.embeddings
 
+        h_attn = attn_output.embeddings
         h_attn = layer_norm(h_attn)
         h += h_attn
         h = with_sharding_constraint(h, sharding)
@@ -1165,15 +1162,17 @@ class LanguageModelConfig:
     _initialized = False
 
     def initialize(self):
-        # We cannot specify [] as a default value (it is mutable), hence None.
         model_config = self.model
-        assert self.init_scale_override is None, (
-            "Overriding model initialize scale is supported only for predefined models."
-        )
+
+        if self.model is None: # We cannot specify [] as a default value (it is mutable), hence None.
+            raise ValueError("Model configuration is not set.")
+        if self.init_scale_override is not None:
+            raise ValueError("Overriding model initialize scale is supported only for predefined models.")
+
         if self.model_size == 0:
             self.model_size = model_config.emb_size
-        assert self.model is not None, "Model could not be initialized."
         self._initialized = True
+
         return self
 
     def make(self, *args, **kwargs):
@@ -1194,7 +1193,7 @@ class LanguageModelConfig:
         return LM_PARTITION_RULES + self.model.partition_rules()
 
 
-def layer_norm(x, model):
+def layer_norm(x):
     return hk_rms_norm(x)
 
 
@@ -1213,17 +1212,12 @@ class LanguageModel(hk.Module):
         tokens: jax.Array,
         memory: Optional[Memory] = None,
         *,
-        batch: Dict[str, jax.Array] = {},
         last_hid_only: bool = False,
         length: Optional[jax.Array] = None,
     ) -> LanguageModelOutput:
         """Forward pass, producing a sequence of logits."""
-        del batch  # Unused.
-
         config = self.config
 
-        input_mask = jnp.greater(tokens, config.pad_token)
-
         # Embed the input tokens and positions.
         in_out_embed = InOutEmbed(
             self.config.vocab_size,
@@ -1235,6 +1229,7 @@ class LanguageModel(hk.Module):
             input_embeddings, P("data", None, self.model.model_axis)
         )
         input_embeddings *= config.embedding_multiplier_scale
+        input_mask = jnp.not_equal(tokens, config.pad_token)
 
         model_output = self.model(
             input_embeddings,
@@ -1242,15 +1237,15 @@ class LanguageModel(hk.Module):
             memory=memory,
         )  # [B, T, D]
         embeddings, model_state = model_output.embeddings, model_output.memory
+        if embeddings.dtype != self.fprop_dtype:
+            raise ValueError(f"Expected forward propagation dtype {self.fprop_dtype} but got {embeddings.dtype} in embeddings.")
+
         if self.model.shard_activations:
-            embeddings = with_sharding_constraint(
-                embeddings, P("data", None, self.model.model_axis)
-            )
+            embeddings = with_sharding_constraint(embeddings, P("data", None, self.model.model_axis))
         else:
             embeddings = with_sharding_constraint(embeddings, P("data", None))
         rank_logger.debug(f"Final embedding shape: {embeddings.shape}")
-        embeddings = layer_norm(embeddings, self.model)
-        assert embeddings.dtype == self.fprop_dtype
+        embeddings = layer_norm(embeddings)
 
         if last_hid_only:
             last_step = jnp.maximum(jnp.sum(input_mask.astype(jnp.int32), axis=1) - 1, 0)

requirements.txt

@@ -1,4 +1,4 @@
 dm_haiku==0.0.12
-jax[cuda12_pip]==0.4.25 -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
+jax[cuda12-pip]==0.4.25 -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
 numpy==1.26.4
 sentencepiece==0.2.0