Fix beam sampler

keras_nlp/samplers/beam_sampler.py

@@ -14,15 +14,13 @@
 """Beam Sampler."""
 
 import tensorflow as tf
-from tensorflow.compiler.tf2xla.python.xla import dynamic_update_slice
 
 from keras_nlp.api_export import keras_nlp_export
 from keras_nlp.backend import keras
 from keras_nlp.backend import ops
 from keras_nlp.samplers.sampler import Sampler
 from keras_nlp.samplers.sampler import call_args_docstring
 from keras_nlp.utils.python_utils import format_docstring
-from keras_nlp.utils.tensor_utils import tensor_to_list
 
 
 @format_docstring(call_args=call_args_docstring)
@@ -53,7 +51,7 @@ class BeamSampler(Sampler):
     batch_size, length, vocab_size = 1, 12, len(int_lookup)
 
     def next(prompt, cache, index):
-        prompt_batch_size = ops.shape(prompt)[0]
+        prompt_batch_size = tf.shape(prompt)[0]
         hidden_states = np.ones((prompt_batch_size, 10))
         # A uniform distribution over our alphabet.
         logits = np.ones((prompt_batch_size, vocab_size))
@@ -76,7 +74,7 @@ def next(prompt, cache, index):
     batch_size, length, vocab_size = 1, 8, len(int_lookup)
 
     def next(prompt, cache, index):
-        prompt_batch_size = ops.shape(prompt)[0]
+        prompt_batch_size = tf.shape(prompt)[0]
         hidden_states = np.ones((prompt_batch_size, 10))
         # A uniform distribution over our alphabet.
         logits = np.ones((batch_size, vocab_size))
@@ -118,39 +116,37 @@ def __call__(
         hidden_states=None,
     ):
         batch_size, max_length = ops.shape(prompt)[0], ops.shape(prompt)[1]
-        # Make sure max length and start index are the same dtype.
-        # index = ops.cast(index, max_length.dtype)
+        index = ops.cast(index, "int32")
 
         def create_beams(x):
             """Add initial beam state."""
             return ops.repeat(x, self.num_beams, axis=0)
 
         def flatten_beams(x):
             """Combine the beam dim and batch dim."""
-            flat_shape = [batch_size * self.num_beams] + tensor_to_list(
-                x.shape
-            )[2:]
-            return ops.reshape(x, new_shape=flat_shape)
+            flat_shape = (batch_size * self.num_beams,) + tuple(x.shape)[2:]
+            return ops.reshape(x, flat_shape)
 
         def unflatten_beams(x):
             """Separate the beam dim and batch dim."""
-            unflat_shape = [batch_size, self.num_beams] + tensor_to_list(
-                x.shape
-            )[1:]
-            return ops.reshape(x, new_shape=unflat_shape)
+            unflat_shape = (batch_size, self.num_beams) + tuple(x.shape)[1:]
+            return ops.reshape(x, unflat_shape)
 
         if mask is None:
             mask = ops.zeros_like(prompt, dtype="bool")
         else:
             mask = ops.cast(mask, dtype="bool")
         # `ops.while_loop` will not accept `None` as a value for `loop_vars`.
-        cache = () if cache is None else cache
+        has_cache = cache is not None
+        cache = cache if has_cache else ()
         # Add extra sequences for each beam.
         prompt, mask = create_beams(prompt), create_beams(mask)
         cache = tf.nest.map_structure(create_beams, cache)
         # Setup the initial beam log-likelihoods.
         # On the first loop, make sure only the original beam is considered.
-        log_probs = [[0.0] + [-1e9] * (self.num_beams - 1)]
+        log_probs = ops.array(
+            [[0.0] + [-1e9] * (self.num_beams - 1)], dtype="float32"
+        )
         log_probs = flatten_beams(ops.repeat(log_probs, batch_size, axis=0))
 
         def cond(prompt, cache, index, log_probs):
@@ -159,7 +155,7 @@ def cond(prompt, cache, index, log_probs):
             # Stop if all sequences have produced a *new* end_token_id.
             end_tokens = (prompt == end_token_id) & (~mask)
             prompt_done = ops.any(end_tokens, axis=-1)
-            return not ops.all(prompt_done)
+            return ops.logical_not(ops.all(prompt_done))
 
         def body(prompt, cache, index, log_probs):
             # Compute the softmax distribution for the next token.
@@ -168,45 +164,47 @@ def body(prompt, cache, index, log_probs):
             probs = keras.activations.softmax(logits / self.temperature)
 
             # Compute the running log-likelihood of each new candidate.
-            next_log_probs = ops.log(probs) + log_probs[..., tf.newaxis]
+            next_log_probs = ops.log(probs) + log_probs[..., None]
             # Reshape `preds` to shape `(batch_size, num_beams * vocab_size)`.
-            next_log_probs = ops.reshape(
-                next_log_probs, new_shape=[batch_size, -1]
-            )
+            next_log_probs = ops.reshape(next_log_probs, [batch_size, -1])
 
             # Compute the top beam indices and next tokens.
             next_log_probs, indices = ops.top_k(
                 next_log_probs, k=self.num_beams, sorted=False
             )
             beam_indices = indices // vocab_size
             next_token = flatten_beams(indices % vocab_size)
-            # Ensure shape is `[None]`, otherwise it causes issues after
-            # converting to TFLite.
-            # next_token = tf.ensure_shape(next_token, [None])
             # We need `ensure_shape` as `top_k` will change the static shape.
             next_log_probs = flatten_beams(next_log_probs)
-            # log_probs = tf.ensure_shape(next_log_probs, log_probs.shape)
+            # Work around tensor shape when not using XLA.
+            if isinstance(log_probs, tf.Tensor):
+                log_probs = tf.ensure_shape(next_log_probs, log_probs.shape)
+            else:
+                log_probs = next_log_probs
 
             def gather_beams(x):
                 x = unflatten_beams(x)
-                print(x.shape, beam_indices.shape)
-                x = ops.take_along_axis(x, beam_indices, axis=1)
+                indices = beam_indices
+                for axis in range(2, len(x.shape)):
+                    indices = ops.expand_dims(indices, axis=axis)
+                x = ops.take_along_axis(x, indices, axis=1)
                 return flatten_beams(x)
 
             prompt = gather_beams(prompt)
-            cache = tf.nest.map_structure(gather_beams, cache)
+            if has_cache:
+                cache = tf.nest.map_structure(gather_beams, cache)
 
             # Update each beam with the next token.
             next_token = ops.cast(next_token, prompt.dtype)
             # Don't overwrite anywhere mask is True.
             next_token = ops.where(mask[:, index], prompt[:, index], next_token)
             # Update the prompt with the next token.
-            next_token = next_token[:, tf.newaxis]
-            prompt = dynamic_update_slice(prompt, next_token, [0, index])
+            next_token = next_token[:, None]
+            prompt = ops.slice_update(prompt, [0, index], next_token)
             # Return the iteration of the loop state.
             return (prompt, cache, index + 1, log_probs)
 
-        prompt, _, _, log_probs = ops.while_loop(
+        prompt, _, _, log_probs = self.run_loop(
             cond=cond,
             body=body,
             loop_vars=(prompt, cache, index, log_probs),
@@ -217,18 +215,21 @@ def gather_beams(x):
         all_log_probs = unflatten_beams(log_probs)
 
         if self.return_all_beams:
-            sorted_indices = ops.argsort(all_log_probs, axis=-1)
-            sorted_indices = ops.flip(sorted_indices, axis=-1)
+            sorted_indices = ops.argsort(-all_log_probs, axis=-1)
             sorted_log_probs = ops.take_along_axis(
-                all_log_probs, sorted_indices, axis=-1
+                all_log_probs,
+                sorted_indices,
+                axis=1,
             )
             sorted_prompts = ops.take_along_axis(
-                all_prompts, sorted_indices, axis=1
+                all_prompts,
+                ops.expand_dims(sorted_indices, -1),
+                axis=1,
             )
             return sorted_prompts, sorted_log_probs
         else:
             # Gather the top beam at each batch index.
-            top_beams = ops.argmax(all_log_probs, axis=-1)[:, tf.newaxis]
+            top_beams = ops.argmax(all_log_probs, axis=-1)[:, None, None]
             prompt = ops.take_along_axis(all_prompts, top_beams, axis=1)
             return ops.squeeze(prompt, axis=1)