Merge pull request #58 from VE-FORBRYDERNE/xmap

Dynamic TPU backend xmaps

aiserver.py

@@ -22,7 +22,7 @@ import packaging
 import contextlib
 import traceback
 import threading
-from typing import Any, Callable, TypeVar, Union, Dict, Set, List
+from typing import Any, Callable, TypeVar, Tuple, Union, Dict, Set, List
 
 import requests
 import html
@@ -993,7 +993,7 @@ else:
                 -1,
                 tpu_mtj_backend.params["d_model"],
             )
-            vars.sp = tensor
+            vars.sp = tpu_mtj_backend.shard_xmap(tensor)
         soft_tokens = np.arange(
             tpu_mtj_backend.params["n_vocab"] + tpu_mtj_backend.params["n_vocab_padding"],
             tpu_mtj_backend.params["n_vocab"] + tpu_mtj_backend.params["n_vocab_padding"] + vars.sp_length,
@@ -1001,6 +1001,49 @@ else:
         )
         return soft_tokens
 
+    def tpumtjgenerate_warper_callback(scores) -> "np.array":
+        scores_shape = scores.shape
+        scores_list = scores.tolist()
+        vars.lua_koboldbridge.logits = vars.lua_state.table()
+        for r, row in enumerate(scores_list):
+            vars.lua_koboldbridge.logits[r+1] = vars.lua_state.table(*row)
+        vars.lua_koboldbridge.vocab_size = scores_shape[-1]
+
+        execute_genmod()
+
+        scores = np.array(
+            tuple(tuple(row.values()) for row in vars.lua_koboldbridge.logits.values()),
+            dtype=scores.dtype,
+        )
+        assert scores.shape == scores_shape
+
+        return scores
+    
+    def tpumtjgenerate_stopping_callback(generated, n_generated, excluded_world_info) -> Tuple[List[set], bool, bool]:
+        vars.generated_tkns += 1
+
+        assert len(excluded_world_info) == len(generated)
+        regeneration_required = vars.lua_koboldbridge.regeneration_required
+        halt = not vars.lua_koboldbridge.generating or vars.generated_tkns >= vars.genamt
+        vars.lua_koboldbridge.regeneration_required = False
+
+        global past
+
+        for i in range(vars.numseqs):
+            vars.lua_koboldbridge.generated[i+1][vars.generated_tkns] = int(generated[i, tpu_mtj_backend.params["seq"] + n_generated - 1].item())
+
+        if(not vars.dynamicscan or halt):
+            return excluded_world_info, regeneration_required, halt
+
+        for i, t in enumerate(generated):
+            decoded = tokenizer.decode(past[i]) + tokenizer.decode(t[tpu_mtj_backend.params["seq"] : tpu_mtj_backend.params["seq"] + n_generated])
+            _, found = checkworldinfo(decoded, force_use_txt=True)
+            found -= excluded_world_info[i]
+            if(len(found) != 0):
+                regeneration_required = True
+                break
+        return excluded_world_info, regeneration_required, halt
+
     # If we're running Colab or OAI, we still need a tokenizer.
     if(vars.model == "Colab"):
         from transformers import GPT2TokenizerFast
@@ -1013,6 +1056,8 @@ else:
         print("{0}Initializing Mesh Transformer JAX, please wait...{1}".format(colors.PURPLE, colors.END))
         assert vars.model == "TPUMeshTransformerGPTJ" and vars.custmodpth and os.path.isdir(vars.custmodpth)
         import tpu_mtj_backend
+        tpu_mtj_backend.warper_callback = tpumtjgenerate_warper_callback
+        tpu_mtj_backend.stopping_callback = tpumtjgenerate_stopping_callback
         tpu_mtj_backend.load_model(vars.custmodpth)
         vars.allowsp = True
         vars.modeldim = int(tpu_mtj_backend.params["d_model"])
@@ -1020,12 +1065,14 @@ else:
         soft_tokens = tpumtjgetsofttokens()
         threading.Thread(  # Compile backend code in background
             target=tpu_mtj_backend.infer,
-            args=(np.uint32((23403, 727, 20185)),),
+            args=(np.tile(np.uint32((23403, 727, 20185)), (vars.numseqs, 1)),),
             kwargs={
                 "soft_embeddings": vars.sp,
                 "soft_tokens": soft_tokens,
+                "use_callback": False,
                 "gen_len": 1,
                 "numseqs": vars.numseqs,
+                "excluded_world_info": list(set() for _ in range(vars.numseqs)),
             },
         ).start()
 
@@ -2890,22 +2937,31 @@ def sendtocolab(txt, min, max):
 #  Send text to TPU mesh transformer backend
 #==================================================================#
 def tpumtjgenerate(txt, minimum, maximum, found_entries=None):
+    vars.generated_tkns = 0
+
     if(found_entries is None):
         found_entries = set()
     found_entries = tuple(found_entries.copy() for _ in range(vars.numseqs))
 
     print("{0}Min:{1}, Max:{2}, Txt:{3}{4}".format(colors.YELLOW, minimum, maximum, tokenizer.decode(txt), colors.END))
 
+    vars._actions = vars.actions
+    vars._prompt = vars.prompt
+    if(vars.dynamicscan):
+        vars._actions = vars._actions.copy()
+
     # Submit input text to generator
     try:
-        if(vars.dynamicscan):
-            raise ValueError("Dynamic world info scanning is not supported by the TPU backend yet")
-
+        context = np.tile(np.uint32(txt), (vars.numseqs, 1))
         soft_tokens = tpumtjgetsofttokens()
 
-        genout = tpool.execute(
+        global past
+        past = np.empty((vars.numseqs, 0), dtype=np.uint32)
+
+        while(True):
+            genout, n_generated, regeneration_required, halt = tpool.execute(
                 tpu_mtj_backend.infer,
-            np.uint32(txt),
+                context,
                 gen_len = maximum-minimum+1,
                 temp=vars.temp,
                 top_p=vars.top_p,
@@ -2915,6 +2971,34 @@ def tpumtjgenerate(txt, minimum, maximum, found_entries=None):
                 repetition_penalty=vars.rep_pen,
                 soft_embeddings=vars.sp,
                 soft_tokens=soft_tokens,
+                excluded_world_info=found_entries,
+            )
+
+            past = np.pad(past, ((0, 0), (0, n_generated)))
+            for r in range(vars.numseqs):
+                for c in range(vars.lua_koboldbridge.generated_cols):
+                    assert vars.lua_koboldbridge.generated[r+1][c+1] is not None
+                    past[r, c] = vars.lua_koboldbridge.generated[r+1][c+1]
+
+            if(halt or not regeneration_required):
+                break
+            print("(regeneration triggered)")
+
+            encoded = []
+            for i in range(vars.numseqs):
+                txt = tokenizer.decode(past[i])
+                winfo, mem, anotetxt, _found_entries = calcsubmitbudgetheader(txt, force_use_txt=True)
+                found_entries[i].update(_found_entries)
+                txt, _, _ = calcsubmitbudget(len(vars._actions), winfo, mem, anotetxt, vars._actions, submission=txt)
+                encoded.append(np.array(txt, dtype=np.uint32))
+            max_length = len(max(encoded, key=len))
+            encoded = np.stack(tuple(np.pad(e, (max_length - len(e), 0), constant_values=tpu_mtj_backend.pad_token_id) for e in encoded))
+            context = np.concatenate(
+                (
+                    encoded,
+                    past,
+                ),
+                axis=-1,
             )
 
     except Exception as e:
@@ -2933,8 +3017,8 @@ def tpumtjgenerate(txt, minimum, maximum, found_entries=None):
         return
 
     for i in range(vars.numseqs):
-        vars.lua_koboldbridge.generated[i+1] = vars.lua_state.table(*genout[i].tolist())
-        vars.lua_koboldbridge.outputs[i+1] = tokenizer.decode(genout[i])
+        vars.lua_koboldbridge.outputs[i+1] = tokenizer.decode(past[i])
+    genout = past
 
     execute_outmod()
     if(vars.lua_koboldbridge.regeneration_required):
@@ -4005,7 +4089,7 @@ def spRequest(filename):
             -1,
             tpu_mtj_backend.params["d_model"],
         )
-        vars.sp = np.float32(tensor)
+        vars.sp = tpu_mtj_backend.shard_xmap(np.float32(tensor))
     else:
         vars.sp = torch.from_numpy(tensor)
 

tpu_mtj_backend.py

@@ -1,5 +1,5 @@
 import multiprocessing
-from typing import Any, Dict, List, Optional
+from typing import Any, Callable, Dict, List, Optional, Tuple, TypeVar
 import progressbar
 import time
 import os
@@ -20,6 +20,13 @@ from mesh_transformer.transformer_shard import CausalTransformer, CausalTransfor
 params: Dict[str, Any] = {}
 
 
+def warper_callback(logits) -> np.array:
+    raise NotImplementedError("`tpu_mtj_backend.warper_callback()` needs to be defined")
+
+def stopping_callback(generated, n_generated, excluded_world_info) -> Tuple[List[set], bool, bool]:
+    raise NotImplementedError("`tpu_mtj_backend.stopping_callback()` needs to be defined")
+
+
 def show_spinner():
     bar = progressbar.ProgressBar(max_value=progressbar.UnknownLength, widgets=[progressbar.Timer(), '  ', progressbar.BouncingBar(left='[', right=']', marker='█')])
     i = 0
@@ -28,6 +35,31 @@ def show_spinner():
         time.sleep(0.1)
         i += 1
 
+
+__F = TypeVar("__F", bound=Callable)
+__T = TypeVar("__T")
+
+def __move_xmap(f: __F, out_axis: str) -> __F:
+    return maps.xmap(
+        f,
+        in_axes=(["shard", ...], ["batch", ...]),
+        out_axes=[out_axis, ...],
+        axis_resources={'shard': 'mp', 'batch': 'dp'},
+    )
+
+def __shard_xmap(batch_dim=1):
+    xmap = __move_xmap(lambda s, b: s, "shard")
+    def inner(x: __T) -> __T:
+        return xmap(x, np.empty(batch_dim))
+    return inner
+
+def __batch_xmap(shard_dim=1):
+    xmap = __move_xmap(lambda s, b: b, "batch")
+    def inner(x: __T) -> __T:
+        return xmap(np.empty(shard_dim), x)
+    return inner
+
+
 def apply_repetition_penalty(logits, tokens, repetition_penalty):
     '''
     This gets called by generate_loop_fn to apply repetition penalty
@@ -38,19 +70,20 @@ def apply_repetition_penalty(logits, tokens, repetition_penalty):
     # logits array; e.g.
     # if logits is [77, 5, 3, 98] and tokens is [0, 1, 2, 3, 2, 3, 1],
     # then penalty_logits will be [77, 5, 3, 98, 3, 98, 5]
-    penalty_logits = jnp.take(logits, tokens)
+    penalty_logits = np.take(logits, tokens)
     # Divide positive values by repetition_penalty and multiply negative
     # values by repetition_penalty (the academic publication that described
     # this technique actually just only divided, but that would cause tokens
     # with negative logits to become more likely, which is obviously wrong)
-    penalty_logits = jnp.where(
+    penalty_logits = np.where(
         penalty_logits > 0,
         penalty_logits/repetition_penalty,
         penalty_logits*repetition_penalty,
     )
     # Finally, put those penalized logit values back into their original
     # positions in the logits array
-    return logits.at[tokens].set(penalty_logits)
+    logits[tokens] = penalty_logits
+    return logits
 
 def kobold_sample(key, logits, top_p=0.9, temp=0.5, top_k=0, tfs=1.0):
     '''
@@ -66,15 +99,16 @@ def kobold_sample(key, logits, top_p=0.9, temp=0.5, top_k=0, tfs=1.0):
         # in the sorted logits array we want to remove and False for ones
         # we want to keep, in this case the first top_k elements will be
         # False and the rest will be True
-        sorted_indices_to_remove = jnp.arange(len(logits)) >= top_k
+        sorted_indices_to_remove = np.arange(len(logits)) >= top_k
         # Unsort the logits array back to its original configuration and
         # remove tokens we need to remove
         _, indices_to_remove = jax.lax.sort_key_val(
-            jnp.argsort(-logits),
+            np.argsort(-logits),
             sorted_indices_to_remove,
         )
-        return jnp.where(indices_to_remove, -jnp.inf, logits)
-    logits = jax.lax.cond(top_k > 0, top_k_filter, lambda x: x, logits)
+        return np.where(indices_to_remove, -np.inf, logits)
+    if top_k > 0:
+        logits = top_k_filter(logits)
     # Top-p (after sorting the remaining tokens again in descending order of
     # logit, remove the ones that have cumulative softmax probability
     # greater than p)
@@ -83,109 +117,167 @@ def kobold_sample(key, logits, top_p=0.9, temp=0.5, top_k=0, tfs=1.0):
         # with e (Euler's number) to the power of that element, and divide
         # each element of the new array by the sum of the elements in the
         # new array
-        sorted_logits = -jnp.sort(-logits)
-        probabilities = jax.nn.softmax(sorted_logits)
+        sorted_logits = -np.sort(-logits)
+        probabilities = np.array(jax.nn.softmax(sorted_logits), copy=True)
         # Calculate cumulative_probabilities as the prefix-sum array of
         # probabilities
-        cumulative_probabilities = jnp.cumsum(probabilities, axis=-1)
+        cumulative_probabilities = np.cumsum(probabilities, axis=-1)
         # We want to remove tokens with cumulative probability higher
         # than top_p
         sorted_indices_to_remove = cumulative_probabilities > top_p
         # Don't ever remove the token with the highest logit, even if
         # the probability is higher than top_p
-        sorted_indices_to_remove = sorted_indices_to_remove.at[0].set(False)
+        sorted_indices_to_remove[0] = False
         # Unsort and remove
         _, indices_to_remove = jax.lax.sort_key_val(
-            jnp.argsort(-logits),
+            np.argsort(-logits),
             sorted_indices_to_remove,
         )
-        return jnp.where(indices_to_remove, -jnp.inf, logits)
-    logits = jax.lax.cond(top_p < 1.0, top_p_filter, lambda x: x, logits)
+        return np.where(indices_to_remove, -np.inf, logits)
+    if top_p < 1.0:
+        logits = top_p_filter(logits)
     # Tail free sampling (basically top-p a second time on remaining tokens
     # except it's the "cumulative normalized absolute second finite
     # differences of the softmax probabilities" instead of just the
     # cumulative softmax probabilities)
     def tail_free_filter(logits):
         # Sort in descending order
-        sorted_logits = -jnp.sort(-logits)
+        sorted_logits = -np.sort(-logits)
         # Softmax again
-        probabilities = jax.nn.softmax(sorted_logits)
+        probabilities = np.array(jax.nn.softmax(sorted_logits), copy=True)
         # Calculate the second finite differences of that array (i.e.
         # calculate the difference array and then calculate the difference
         # array of the difference array)
-        d2 = jnp.diff(jnp.diff(probabilities))
+        d2 = np.diff(np.diff(probabilities))
         # Get the absolute values of all those second finite differences
-        d2 = jnp.abs(d2)
+        d2 = np.abs(d2)
         # Normalize (all elements in the array are divided by the sum of the
         # array's elements)
         d2 = d2 / d2.sum(axis=-1, keepdims=True)
         # Get the prefix-sum array
-        cumulative_d2 = jnp.cumsum(d2, axis=-1)
+        cumulative_d2 = np.cumsum(d2, axis=-1)
         # We will remove the tokens with a cumulative normalized absolute
         # second finite difference larger than the TFS value
         sorted_indices_to_remove = cumulative_d2 > tfs
         # Don't remove the token with the highest logit
-        sorted_indices_to_remove = sorted_indices_to_remove.at[0].set(False)
+        sorted_indices_to_remove[0] = False
         # Since the d2 array has two fewer elements than the logits array,
         # we'll add two extra Trues to the end
-        sorted_indices_to_remove = jnp.pad(
+        sorted_indices_to_remove = np.pad(
             sorted_indices_to_remove,
             (0, 2),
             constant_values=True,
         )
         # Unsort and remove
         _, indices_to_remove = jax.lax.sort_key_val(
-            jnp.argsort(-logits),
+            np.argsort(-logits),
             sorted_indices_to_remove,
         )
-        return jnp.where(indices_to_remove, -jnp.inf, logits)
-    logits = jax.lax.cond(tfs < 1.0, tail_free_filter, lambda x: x, logits)
+        return np.where(indices_to_remove, -np.inf, logits)
+    if tfs < 1.0:
+        logits = tail_free_filter(logits)
     # Temperature (just divide the logits by the temperature)
-    def temp_filter(logits):
-        return logits / temp
-    logits = jax.lax.cond(True, temp_filter, lambda x: x, logits)
+    logits /= temp
     # Finally, pick one token using the softmax thingy again (it gives
     # an array whose elements sum to 1 so it can be used nicely as a
     # probability distribution)
-    return jax.random.categorical(key, logits, -1).astype(jnp.uint32)[jnp.newaxis]
+    return jax.random.categorical(key, logits, -1).astype(np.uint32)
 
 pad_token_id = 50256
 
+def sample_func(data, key, numseqs_aux, badwords, repetition_penalty, sampler_options):
+    numseqs = numseqs_aux.shape[0]
+    gi = data[0][1]
+    def sample_loop_fn(carry):
+        generated, generated_index, logits, _ = carry[0][0]
+        sample_key = carry[1]
+        # Get the pseudo-random number generator key that will
+        # be used by kobold_sample to randomly pick a token
+        sample_key, new_key = jax.random.split(sample_key, num=2)
+        # Apply repetition penalty to all tokens that are
+        # currently inside the "generated" array
+        logits = apply_repetition_penalty(
+            logits,
+            generated,
+            repetition_penalty
+        )
+        # Remove any tokens in the badwords list by setting
+        # their logits to negative infinity which effectively
+        # makes their probabilities of being chosen zero
+        logits[badwords] = -np.inf
+        # Use the sampler (kobold_sample) to pick one token
+        # based on the logits array as a 0D uint32 array
+        # (higher logit means higher probability of being
+        # picked, non-linearly)
+        next_token = kobold_sample(
+            sample_key,
+            logits,
+            **sampler_options,
+        )
+        # Remember what token was picked
+        generated[generated_index] = next_token
+        generated_index += 1
+        # Re-pack the current sample_loop_fn's state so we can
+        # get back the same variables the next time
+        carry[0][0] = [generated, generated_index, logits, next_token]
+        carry[0].append(carry[0].pop(0))
+        return carry[0], new_key
+    # return jax.lax.while_loop(
+    #     lambda carry: carry[0][0][1] == gi,
+    #     sample_loop_fn,
+    #     (data, key),
+    # )
+    carry = (data, key)
+    while carry[0][0][1] == gi:
+        carry = sample_loop_fn(carry)
+    return carry
+
 class PenalizingCausalTransformer(CausalTransformer):
     def __init__(self, config):
         # Initialize
         super().__init__(config)
-        def generate(state, key, ctx, ctx_length, gen_length, numseqs_aux, sampler_options, soft_embeddings=None):
+        def generate_initial(state, key, ctx, ctx_length, numseqs_aux, soft_embeddings=None):
             numseqs = numseqs_aux.shape[0]
-            # These are the tokens that we don't want the AI to ever write
-            self.badwords = jnp.array([6880, 50256, 42496, 4613, 17414, 22039, 16410, 27, 29, 38430, 37922, 15913, 24618, 28725, 58, 47175, 36937, 26700, 12878, 16471, 37981, 5218, 29795, 13412, 45160, 3693, 49778, 4211, 20598, 36475, 33409, 44167, 32406, 29847, 29342, 42669, 685, 25787, 7359, 3784, 5320, 33994, 33490, 34516, 43734, 17635, 24293, 9959, 23785, 21737, 28401, 18161, 26358, 32509, 1279, 38155, 18189, 26894, 6927, 14610, 23834, 11037, 14631, 26933, 46904, 22330, 25915, 47934, 38214, 1875, 14692, 41832, 13163, 25970, 29565, 44926, 19841, 37250, 49029, 9609, 44438, 16791, 17816, 30109, 41888, 47527, 42924, 23984, 49074, 33717, 31161, 49082, 30138, 31175, 12240, 14804, 7131, 26076, 33250, 3556, 38381, 36338, 32756, 46581, 17912, 49146])
-            def generate_sample(context, ctx_length):
+            @hk.transform
+            def generate_initial_inner(context, ctx_length):
                 # Give the initial context to the transformer
                 transformer = CausalTransformerShard(config)
-                def generate_initial_scan_fn(sequence_index, _):
-                    _, initial_state = transformer.generate_initial(context, ctx_length, soft_embeddings=soft_embeddings)
-                    # The "generated" array will contain the tokens from the
-                    # context as well as the tokens picked by the sampler at
-                    # each stage, padded with a bunch of 50256s, so we know
-                    # which tokens have to be repetition penalized
-                    generated = jnp.pad(context, (0, config["seq"]), constant_values=pad_token_id)  # Let it start off with just the 2048 context tokens, plus some 50256s which will be eventually filled with sampler-chosen tokens
+                def generate_initial_scan_fn(sequence_index, c):
+                    _, initial_state = transformer.generate_initial(c, ctx_length, soft_embeddings=soft_embeddings)
                     generated_index = config["seq"]
                     # Add that information to generate_loop_fn's starting state
-                    initial_state = (generated, generated_index, sequence_index) + initial_state
+                    initial_state = (jnp.empty(config["n_vocab"], dtype=jnp.float32), generated_index, sequence_index) + initial_state
                     return sequence_index+1, initial_state
-                _, initial_states = jax.lax.scan(generate_initial_scan_fn, 0, None, numseqs)
+                _, initial_states = jax.lax.scan(generate_initial_scan_fn, 0, context, numseqs)
                 sample_key = initial_states[-1][0]
-                initial_states = list(jax.tree_map(lambda x: x[i], initial_states[:-1]) for i in range(numseqs))
-                # Get repetition penalty from the arguments
-                repetition_penalty = sampler_options.pop('repetition_penalty', None)
+                initial_states = list(list(jax.tree_map(lambda x: x[i], initial_states[:-1])) for i in range(numseqs))
+                return initial_states, sample_key
+            return generate_initial_inner.apply(state["params"], key, ctx, ctx_length)
+        self.generate_initial_xmap = jax.experimental.maps.xmap(
+            fun=generate_initial,
+            in_axes=(
+                ["shard", ...],
+                ["batch", ...],
+                ["batch", ...],
+                ["batch", ...],
+                ["batch", ...],
+                ["shard", ...],
+            ),
+            out_axes=["shard", "batch", ...],
+            axis_resources={'shard': 'mp', 'batch': 'dp'},
+        )
+        def generate_once(data, state, numseqs_aux, soft_embeddings=None):
+            numseqs = numseqs_aux.shape[0]
+            @hk.without_apply_rng
+            @hk.transform
+            def generate_once_inner():
+                gi = data[0][1]
+                # Give the initial context to the transformer
+                transformer = CausalTransformerShard(config)
                 # This is the main generation loop
                 def generate_loop_fn(carry):
                     # Unpack current generate_loop_fn state
-                    generated, generated_index, sequence_index, next_token, decode_state = carry[0][0]
-                    sample_key = carry[1]
-                    # Get the pseudo-random number generator key that will
-                    # be used by kobold_sample to randomly pick a token
-                    sample_key, new_key = jax.random.split(sample_key)
+                    _, generated_index, sequence_index, next_token, decode_state = carry[0][0]
                     # Give the context to the model and get the logits it
                     # spits out
                     # (a 2D array with 1 row and 50400 columns representing
@@ -196,75 +288,78 @@ class PenalizingCausalTransformer(CausalTransformer):
                     # Verify that logits does indeed have that many rows and
                     # columns (if you get an error here, pray for mercy)
                     assert logits.shape == (1, config["n_vocab"])
+                    assert logits.dtype == jnp.float32
                     # Flatten it into a 1D array to make it easier to use
                     logits = logits[0]
-                    # Apply repetition penalty to all tokens that are
-                    # currently inside the "generated" array
-                    if repetition_penalty is not None:
-                        logits = apply_repetition_penalty(
-                            logits,
-                            generated,
-                            repetition_penalty
-                        )
-                    # Remove any tokens in the badwords list by setting
-                    # their logits to negative infinity which effectively
-                    # makes their probabilities of being chosen zero
-                    logits = logits.at[self.badwords].set(-jnp.inf)
-                    # Use the sampler (kobold_sample) to pick one token
-                    # based on the logits array as a 1D array with 1 element
-                    # (higher logit means higher probability of being
-                    # picked, non-linearly)
-                    next_token = kobold_sample(
-                        sample_key,
-                        logits,
-                        **sampler_options,
-                    )
-                    # Remember what token was picked
-                    generated = generated.at[generated_index].set(next_token[0])
-                    generated_index += 1
                     # Re-pack the current generate_loop_fn's state so we can
                     # get back the same variables the next time
-                    carry[0][0] = (generated, generated_index, sequence_index, next_token, new_state)
+                    generated_index += 1
+                    carry[0][0] = [logits, generated_index, sequence_index, next_token, new_state]
                     carry[0].append(carry[0].pop(0))
-                    return carry[0], new_key
-                final_state = jax.lax.while_loop(
-                    lambda carry: carry[0][0][1] - config["seq"] < gen_length,
+                    return carry[0],
+                return jax.lax.while_loop(
+                    lambda carry: carry[0][0][1] == gi,
                     generate_loop_fn,
-                    (initial_states, sample_key),
+                    (data,),
                 )
-                return final_state
-            generate_fn = hk.transform(generate_sample).apply
-            return generate_fn(state["params"], key, ctx, ctx_length)
-        self.generate_xmap = jax.experimental.maps.xmap(
-            fun=generate,
+            return generate_once_inner.apply(state["params"])
+        self.generate_once_xmap = jax.experimental.maps.xmap(
+            fun=generate_once,
             in_axes=(
+                ["shard", "batch", ...],
                 ["shard", ...],
                 ["batch", ...],
-                ["batch", ...],
-                ["batch", ...],
-                ["batch", ...],
-                ["batch", ...],
-                ["batch", ...],
                 ["shard", ...],
             ),
             out_axes=["shard", "batch", ...],
             axis_resources={'shard': 'mp', 'batch': 'dp'},
         )
-    def generate(self, ctx, ctx_length, gen_length, numseqs, sampler_options, return_logits=False, soft_embeddings=None):
+    def generate(self, ctx, ctx_length, gen_length, numseqs, sampler_options, return_logits=False, soft_embeddings=None, excluded_world_info=None, use_callback=True):
+        assert excluded_world_info is not None
         assert not return_logits
+        assert gen_length.ndim == 1
+        assert soft_embeddings is not None
         key = hk.PRNGSequence(random.randint(0, 2 ** 60))
         batch_size = ctx.shape[0]
         self.batch_size = batch_size
-        return self.generate_xmap(
-            self.state,
-            jnp.array(key.take(batch_size)),
-            ctx,
-            np.array(ctx_length, dtype=np.uint32),
-            np.array(gen_length, dtype=np.uint32),
-            np.empty((batch_size, numseqs), dtype=np.uint8),
-            sampler_options,
-            soft_embeddings,
-        )
+        _numseqs_aux = jnp.empty((batch_size, numseqs), dtype=np.uint32)
+        numseqs_aux = batch_xmap(_numseqs_aux)
+        sample_data = [
+            [
+                np.pad(ctx[0][i], (0, params["seq"]), constant_values=pad_token_id),
+                params["seq"],
+                None,
+                np.empty((), dtype=np.uint32),
+            ]
+            for i in range(numseqs)
+        ]
+        repetition_penalty = sampler_options.pop("repetition_penalty", 1.0)
+        n_generated = 0
+        regeneration_required = False
+        halt = False
+        generate_data, sample_key = self.generate_initial_xmap(self.state, jnp.array(key.take(batch_size)), ctx, ctx_length, numseqs_aux, soft_embeddings)
+        sample_key = np.asarray(sample_key[0, 0])
+        while True:
+            generate_data, = self.generate_once_xmap(generate_data, self.state, numseqs_aux, soft_embeddings)
+            for i in range(numseqs):
+                sample_data[i][2] = np.array(generate_data[i][0][0, 0], copy=True)
+            if use_callback:
+                logits = np.float32(tuple(d[2] for d in sample_data))
+                logits = warper_callback(logits)
+                for i in range(numseqs):
+                    sample_data[i][2] = logits[i]
+            sample_data, sample_key = sample_func(sample_data, sample_key, _numseqs_aux, badwords, repetition_penalty, sampler_options)
+            n_generated += 1
+            for i in range(numseqs):
+                generate_data[i][3] = np.tile(sample_data[i][0][sample_data[i][1]-1][np.newaxis, np.newaxis], (params["cores_per_replica"], 1, 1))
+            if use_callback:
+                generated = np.uint32(tuple(d[0] for d in sample_data))
+                excluded_world_info, regeneration_required, halt = stopping_callback(generated, n_generated, excluded_world_info)
+                if regeneration_required or halt:
+                    break
+            else:
+                break
+        return sample_data, n_generated, regeneration_required, halt
 
 
 def infer(
@@ -278,35 +373,40 @@ def infer(
     gen_len=80,
     soft_embeddings: Optional[np.array] = None,
     soft_tokens: Optional[np.array] = None,
-) -> List[str]:
+    excluded_world_info = None,
+    use_callback=True,
+) -> Tuple[List[np.array], int, bool, bool]:
+    assert excluded_world_info is not None
     maps.thread_resources.env = thread_resources_env
     total_batch = 1
     tokens = context
     if(soft_tokens is not None):
-        tokens = np.uint32(np.concatenate((soft_tokens, tokens)))
-    provided_ctx = tokens.shape[0]
+        tokens = np.uint32(np.concatenate((np.tile(soft_tokens, (tokens.shape[0], 1)), tokens), axis=-1))
+    provided_ctx = tokens.shape[-1]
     pad_amount = seq - provided_ctx
-    padded_tokens = np.pad(tokens, ((pad_amount, 0),), constant_values=pad_token_id)
+    padded_tokens = np.pad(tokens, ((0, 0), (pad_amount, 0)), constant_values=pad_token_id)
     batched_tokens = np.array([padded_tokens] * total_batch)
     samples = []
-    batched_generator_params = {
-        "temp": temp * np.ones(total_batch),
-        "top_p": top_p * np.ones(total_batch),
-        "tfs": tfs * np.ones(total_batch),
-        "repetition_penalty": repetition_penalty * np.ones(total_batch),
-        "top_k": np.full(total_batch, top_k, dtype=np.uint32)
+    generator_params = {
+        "temp": float(temp),
+        "top_p": float(top_p),
+        "tfs": float(tfs),
+        "repetition_penalty": float(repetition_penalty),
+        "top_k": int(top_k),
     }
     output = network.generate(
         batched_tokens,
         np.ones(total_batch, dtype=np.uint32) * provided_ctx,
         np.ones(total_batch, dtype=np.uint32) * gen_len,
         numseqs,
-        batched_generator_params,
+        generator_params,
         soft_embeddings=soft_embeddings,
-    )[0]
-    for o in output:
-        samples.append(o[0][0, 0, params["seq"] : params["seq"] + gen_len])
-    return samples
+        excluded_world_info=excluded_world_info,
+        use_callback=use_callback,
+    )
+    for out in output[0]:
+        samples.append(out[0][params["seq"] : params["seq"] + gen_len])
+    return (samples,) + output[1:]
 
 
 def load_model(path: str, driver_version="tpu_driver0.1_dev20210607", **kwargs) -> None:
@@ -354,6 +454,14 @@ def load_model(path: str, driver_version="tpu_driver0.1_dev20210607", **kwargs)
     maps.thread_resources.env = thread_resources_env
     tokenizer = transformers.GPT2TokenizerFast.from_pretrained('gpt2')
 
+    global shard_xmap, batch_xmap
+    shard_xmap = __shard_xmap()
+    batch_xmap = __batch_xmap(shard_dim=cores_per_replica)
+
+    global badwords
+    # These are the tokens that we don't want the AI to ever write
+    badwords = jnp.array([6880, 50256, 42496, 4613, 17414, 22039, 16410, 27, 29, 38430, 37922, 15913, 24618, 28725, 58, 47175, 36937, 26700, 12878, 16471, 37981, 5218, 29795, 13412, 45160, 3693, 49778, 4211, 20598, 36475, 33409, 44167, 32406, 29847, 29342, 42669, 685, 25787, 7359, 3784, 5320, 33994, 33490, 34516, 43734, 17635, 24293, 9959, 23785, 21737, 28401, 18161, 26358, 32509, 1279, 38155, 18189, 26894, 6927, 14610, 23834, 11037, 14631, 26933, 46904, 22330, 25915, 47934, 38214, 1875, 14692, 41832, 13163, 25970, 29565, 44926, 19841, 37250, 49029, 9609, 44438, 16791, 17816, 30109, 41888, 47527, 42924, 23984, 49074, 33717, 31161, 49082, 30138, 31175, 12240, 14804, 7131, 26076, 33250, 3556, 38381, 36338, 32756, 46581, 17912, 49146])
+
     if not path.endswith("/"):
         path += "/"