feat(server): Support generic AutoModelForCausalLM

README.md

@@ -18,6 +18,7 @@ A Rust and gRPC server for text generation inference.
 ## Supported models
 
 - BLOOM
+- BLOOMZ
 - BLOOM-560m
 
 ## Load Tests for BLOOM

proto/generate.proto

@@ -63,6 +63,8 @@ message GeneratedText {
     Request request = 1;
     /// Output
     string output = 2;
+    /// Number of generated tokens
+    uint32 tokens = 3;
 }
 
 message GenerateRequest {

router/src/batcher.rs

@@ -190,6 +190,7 @@ fn send_generated(finished: Vec<GeneratedText>, db: &Db) {
             .expect("ID not found in db. This is a bug.");
         let response = InferResponse {
             output: output.output,
+            tokens: output.tokens,
             queued: entry.time,
             start: entry.batch_time.unwrap(), // unwrap is always valid
             end: Instant::now(),
@@ -202,6 +203,7 @@ fn send_generated(finished: Vec<GeneratedText>, db: &Db) {
 #[derive(Debug)]
 pub(crate) struct InferResponse {
     pub(crate) output: String,
+    pub(crate) tokens: u32,
     pub(crate) queued: Instant,
     pub(crate) start: Instant,
     pub(crate) end: Instant,

router/src/server.rs

@@ -116,7 +116,7 @@ async fn generate(
     let validation_time = response.queued - start_time;
     let queue_time = response.start - response.queued;
     let inference_time = response.end - response.start;
-    let time_per_token = inference_time / req.parameters.max_new_tokens;
+    let time_per_token = inference_time / response.tokens;
 
     // Headers
     let mut headers = HeaderMap::new();

server/text_generation/models/__init__.py

@@ -1,7 +1,8 @@
 from text_generation.models.model import Model
-from text_generation.models.bloom import BLOOM, BLOOMSharded
+from text_generation.models.bloom import BLOOMSharded
+from text_generation.models.causal_lm import CausalLM
 
-__all__ = ["Model", "BLOOM", "BLOOMSharded"]
+__all__ = ["Model", "BLOOMSharded", "CausalLM"]
 
 
 def get_model(model_name: str, sharded: bool, quantize: bool) -> Model:
@@ -11,6 +12,10 @@ def get_model(model_name: str, sharded: bool, quantize: bool) -> Model:
         else:
             if quantize:
                 raise ValueError("quantization is not supported for non-sharded BLOOM")
-            return BLOOM(model_name)
+            return CausalLM(model_name)
     else:
-        raise ValueError(f"model {model_name} is not supported yet")
+        if sharded:
+            raise ValueError("sharded is not supported for AutoModel")
+        if quantize:
+            raise ValueError("quantize is not supported for AutoModel")
+        return CausalLM(model_name)

server/text_generation/models/bloom.py

@@ -1,7 +1,7 @@
 import torch
 import torch.distributed
 
-from typing import List, Optional, Tuple, Type
+from typing import List, Optional
 
 from accelerate import init_empty_weights
 from safetensors import safe_open
@@ -11,10 +11,8 @@ from transformers.models.bloom.parallel_layers import (
     TensorParallelEmbedding,
     TensorParallelRowLinear,
 )
-from transformers.modeling_outputs import CausalLMOutputWithPast
 
 from text_generation.models import Model
-from text_generation.models.types import Batch, GeneratedText
 from text_generation.utils import (
     initialize_torch_distributed,
     weight_files,
@@ -31,322 +29,26 @@ except Exception as e:
 torch.manual_seed(0)
 
 
-class BloomBatch(Batch):
+class BLOOMSharded(Model):
-    @classmethod
-    def concatenate(cls, batches: List["Batch"]) -> "BloomBatch":
-        # Used for padding
-        total_batch_size = sum(batch.size for batch in batches)
-        max_sequence_length = max(batch.max_sequence_length for batch in batches)
-
-        # Batch attributes
-        input_ids = {"input_ids": None, "attention_mask": None, "past_key_values": []}
-        requests = []
-        all_input_lengths = []
-        all_input_ids = []
-        next_token_choosers = []
-        stopping_criterias = []
-
-        # Used for slicing correctly inside the tensors
-        # Equivalent to a cumsum on batch sizes
-        start_index = 0
-        for i, batch in enumerate(batches):
-            requests.extend(batch.requests)
-            all_input_lengths.extend(batch.all_input_lengths)
-            all_input_ids.extend(batch.all_input_ids)
-            next_token_choosers.extend(batch.next_token_choosers)
-            stopping_criterias.extend(batch.stopping_criterias)
-
-            # Slicing end index for this batch
-            end_index = start_index + batch.size
-
-            # We only concatenate batches that did at least one step
-            if batch.input_ids["input_ids"].shape[1] > 1:
-                raise ValueError("Batch input_ids should be of shape (batch_size, 1)")
-
-            # Initialize tensors
-            if i == 0:
-                input_ids["input_ids"] = torch.empty(
-                    (total_batch_size, 1),
-                    dtype=batch.input_ids["input_ids"].dtype,
-                    device=batch.input_ids["input_ids"].device,
-                )
-                input_ids["attention_mask"] = torch.zeros(
-                    (total_batch_size, max_sequence_length),
-                    dtype=batch.input_ids["attention_mask"].dtype,
-                    device=batch.input_ids["attention_mask"].device,
-                )
-
-            # input_ids["input_ids"] is always of shape [batch_size, 1]
-            # We do not need to pad it
-            input_ids["input_ids"][start_index:end_index] = batch.input_ids["input_ids"]
-
-            # We need to slice the attention mask to remove padding from previous steps
-            input_ids["attention_mask"][
-            start_index:end_index, -batch.max_sequence_length:
-            ] = batch.input_ids["attention_mask"][:, -batch.max_sequence_length:]
-
-            for j, past in enumerate(batch.input_ids["past_key_values"]):
-                past_keys = past[0]
-                past_values = past[1]
-
-                _, head_dim, padded_sequence_length = past_keys.shape
-
-                # Reshape the tensors to make slicing easier
-                past_keys = past_keys.view(
-                    batch.size, -1, head_dim, padded_sequence_length
-                )
-                past_values = past_values.view(
-                    batch.size, -1, padded_sequence_length, head_dim
-                )
-                num_heads = past_keys.shape[1]
-
-                # Initialize tensors
-                # This will run only once per layer
-                if j == len(input_ids["past_key_values"]):
-                    padded_past_keys = torch.zeros(
-                        (
-                            total_batch_size,
-                            num_heads,
-                            head_dim,
-                            max_sequence_length - 1,
-                        ),
-                        dtype=past_keys.dtype,
-                        device=past_keys.device,
-                    )
-                    padded_past_values = torch.zeros(
-                        (
-                            total_batch_size,
-                            num_heads,
-                            max_sequence_length - 1,
-                            head_dim,
-                        ),
-                        dtype=past_values.dtype,
-                        device=past_values.device,
-                    )
-                    input_ids["past_key_values"].append(
-                        [padded_past_keys, padded_past_values]
-                    )
-
-                # We slice the past keys and values to remove the padding from previous batches
-                input_ids["past_key_values"][j][0][
-                start_index:end_index, :, :, -(batch.max_sequence_length - 1):
-                ] = past_keys[:, :, :, -(batch.max_sequence_length - 1):]
-
-                input_ids["past_key_values"][j][1][
-                start_index:end_index, :, -(batch.max_sequence_length - 1):, :
-                ] = past_values[:, :, -(batch.max_sequence_length - 1):, :]
-
-                # If we are on the last batch, we need to reshape the tensors
-                if (i + 1) == len(batches):
-                    input_ids["past_key_values"][j][0] = input_ids["past_key_values"][
-                        j
-                    ][0].view(total_batch_size * num_heads, head_dim, -1)
-                    input_ids["past_key_values"][j][1] = input_ids["past_key_values"][
-                        j
-                    ][1].view(total_batch_size * num_heads, -1, head_dim)
-
-            start_index += batch.size
-
-        return cls(
-            batch_id=batches[0].batch_id,
-            requests=requests,
-            all_input_lengths=all_input_lengths,
-            input_ids=input_ids,
-            all_input_ids=all_input_ids,
-            next_token_choosers=next_token_choosers,
-            stopping_criterias=stopping_criterias,
-            size=total_batch_size,
-            max_sequence_length=max_sequence_length,
-        )
-
-
-class BLOOM(Model):
-    def __init__(self, model_name: str):
-        if not model_name.startswith("bigscience/bloom"):
-            raise ValueError(f"Model {model_name} is not supported")
-
-        if torch.cuda.is_available():
-            self.device = torch.device("cuda")
-            dtype = torch.bfloat16 if torch.cuda.is_bf16_supported() else torch.float16
-        else:
-            self.device = torch.device("cpu")
-            dtype = torch.float32
-
-        self.tokenizer = AutoTokenizer.from_pretrained(model_name, padding_side="left")
-        self.tokenizer.add_special_tokens({"pad_token": "[PAD]"})
-        self.model = AutoModelForCausalLM.from_pretrained(
-            model_name, torch_dtype=dtype, device_map="auto" if torch.cuda.is_available() else None
-        ).eval()
-
-        self.num_heads = self.model.config.num_attention_heads
-
-    @property
-    def batch_type(self) -> Type[BloomBatch]:
-        return BloomBatch
-
-    def forward(
-            self, input_ids, attention_mask, past_key_values: Optional = None
-    ) -> CausalLMOutputWithPast:
-        # Model Forward
-        return self.model.forward(
-            input_ids=input_ids,
-            attention_mask=attention_mask,
-            past_key_values=past_key_values,
-            use_cache=True,
-        )
-
-    def generate_token(
-            self, batch: BloomBatch
-    ) -> Tuple[List[GeneratedText], Optional[BloomBatch]]:
-        # For some reason, inference_mode does not work well with GLOO which we use on CPU
-        context_manager = (
-            torch.no_grad if self.device.type == "cpu" else torch.inference_mode
-        )
-        with context_manager():
-            outputs = self.forward(**batch.input_ids)
-
-        # List of indices to cache
-        next_batch_keep_indices = []
-        next_batch_past_keep_indices = []
-
-        # New input_ids for next forward
-        next_batch_input_ids = []
-        next_batch_all_input_ids = []
-        next_all_input_lengths = []
-
-        next_batch_size = 0
-        next_batch_max_sequence_length = 0
-
-        # Finished requests
-        generated_texts: List[GeneratedText] = []
-
-        # Zipped iterator
-        iterator = zip(
-            batch.requests,
-            batch.all_input_lengths,
-            outputs.logits,
-            batch.next_token_choosers,
-            batch.stopping_criterias,
-            batch.all_input_ids,
-        )
-
-        # For each member of the batch
-        for i, (
-                request,
-                input_length,
-                logits,
-                next_token_chooser,
-                stopping_criteria,
-                all_tokens,
-        ) in enumerate(iterator):
-            # Select next token
-            next_token = next_token_chooser(all_tokens, logits.unsqueeze(0)[:, -1])
-
-            # Append next token to all tokens
-            all_tokens = torch.cat([all_tokens, next_token])
-
-            # Evaluate stopping criteria
-            if stopping_criteria(all_tokens):
-                # Decode all tokens
-                output = self.tokenizer.decode(
-                    all_tokens.squeeze(-1), skip_special_tokens=True
-                )
-                # Add to the list of finished generations with the original request
-                generated_texts.append(GeneratedText(request, output))
-            # add to the next batch
-            else:
-                next_batch_keep_indices.append(i)
-                # past_key_values is of shape [batch_size * num_heads, ...]
-                # so we need to take into account the `num_heads` stride here
-                next_batch_past_keep_indices.extend(
-                    [j for j in range(i * self.num_heads, (i + 1) * self.num_heads)]
-                )
-                next_batch_input_ids.append(next_token)
-                next_batch_all_input_ids.append(all_tokens)
-                next_batch_size += 1
-                new_input_length = input_length + 1
-                next_all_input_lengths.append(new_input_length)
-                next_batch_max_sequence_length = max(
-                    next_batch_max_sequence_length, new_input_length
-                )
-
-        # We finished all generations in the batch; there is no next batch
-        if not next_batch_keep_indices:
-            return generated_texts, None
-
-        # If we finished at least one generation
-        next_batch_input_ids = {"input_ids": torch.cat(next_batch_input_ids, dim=0)}
-        if generated_texts:
-            # Apply indices to attention mask, past key values and other items that need to be cached
-            next_batch_input_ids["attention_mask"] = batch.input_ids["attention_mask"][
-                next_batch_keep_indices
-            ]
-            next_batch_input_ids["past_key_values"] = [
-                (
-                    keys[next_batch_past_keep_indices],
-                    values[next_batch_past_keep_indices],
-                )
-                for keys, values in outputs["past_key_values"]
-            ]
-            next_batch_requests = [batch.requests[i] for i in next_batch_keep_indices]
-            next_batch_next_token_choosers = [
-                batch.next_token_choosers[i] for i in next_batch_keep_indices
-            ]
-            next_batch_stopping_criterias = [
-                batch.stopping_criterias[i] for i in next_batch_keep_indices
-            ]
-        else:
-            next_batch_input_ids["attention_mask"] = batch.input_ids["attention_mask"]
-            next_batch_input_ids["past_key_values"] = outputs["past_key_values"]
-            next_batch_requests = batch.requests
-            next_batch_next_token_choosers = batch.next_token_choosers
-            next_batch_stopping_criterias = batch.stopping_criterias
-
-        # Update attention_mask with padding as we added a new token to input_ids
-        next_batch_input_ids["attention_mask"] = torch.cat(
-            [
-                next_batch_input_ids["attention_mask"],
-                torch.ones((next_batch_size, 1)).to(self.device),
-            ],
-            dim=1,
-        )
-
-        next_batch = BloomBatch(
-            batch_id=batch.batch_id,
-            requests=next_batch_requests,
-            all_input_lengths=next_all_input_lengths,
-            input_ids=next_batch_input_ids,
-            all_input_ids=next_batch_all_input_ids,
-            next_token_choosers=next_batch_next_token_choosers,
-            stopping_criterias=next_batch_stopping_criterias,
-            size=next_batch_size,
-            max_sequence_length=next_batch_max_sequence_length,
-        )
-        return generated_texts, next_batch
-
-
-class BLOOMSharded(BLOOM):
     def __init__(self, model_name: str, quantize: bool = False):
-        super(Model, self).__init__()
         if not model_name.startswith("bigscience/bloom"):
             raise ValueError(f"Model {model_name} is not supported")
 
         self.process_group, self.rank, self.world_size = initialize_torch_distributed()
         self.master = self.rank == 0
         if torch.cuda.is_available():
-            self.device = torch.device(f"cuda:{self.rank}")
+            device = torch.device(f"cuda:{self.rank}")
             dtype = torch.float16
         else:
-            self.device = torch.device("cpu")
+            device = torch.device("cpu")
             dtype = torch.float32
 
-        self.tokenizer = AutoTokenizer.from_pretrained(model_name, padding_side="left")
+        tokenizer = AutoTokenizer.from_pretrained(model_name, padding_side="left")
 
         config = AutoConfig.from_pretrained(
             model_name, slow_but_exact=False, tp_parallel=True
         )
         config.pad_token_id = 3
-        self.num_heads = config.n_head // self.process_group.size()
 
         # The flag below controls whether to allow TF32 on matmul. This flag defaults to False
         # in PyTorch 1.12 and later.
@@ -370,12 +72,14 @@ class BLOOMSharded(BLOOM):
             model,
             filenames,
             quantize=quantize,
-            device=self.device,
+            device=device,
             rank=self.rank,
             world_size=self.world_size,
         )
         self.model = model.eval().to(dtype)
         torch.distributed.barrier(group=self.process_group)
+        super(BLOOMSharded, self).__init__(tokenizer=tokenizer, num_heads=config.n_head // self.process_group.size(),
+                                           device=device)
 
     @staticmethod
     def load_weights(
@@ -526,5 +230,4 @@ class BLOOMSharded(BLOOM):
         torch.distributed.all_gather(logits, logits_shard, group=self.process_group)
         logits = torch.cat(logits, dim=1).view(batch_size, 1, vocab_size)
 
-        outputs.logits = logits
+        return logits, outputs.past_key_values
-        return outputs

server/text_generation/models/causal_lm.py

@@ -0,0 +1,38 @@
+import torch
+
+from transformers import AutoTokenizer, AutoModelForCausalLM
+from typing import Optional, Tuple, List
+
+from text_generation.models import Model
+
+
+class CausalLM(Model):
+    def __init__(self, model_name: str):
+        if torch.cuda.is_available():
+            device = torch.device("cuda")
+            dtype = torch.bfloat16 if torch.cuda.is_bf16_supported() else torch.float32
+        else:
+            device = torch.device("cpu")
+            dtype = torch.float32
+
+        tokenizer = AutoTokenizer.from_pretrained(model_name, padding_side="left")
+        tokenizer.add_special_tokens({"pad_token": "[PAD]"})
+        self.model = AutoModelForCausalLM.from_pretrained(
+            model_name,
+            torch_dtype=dtype,
+            device_map="auto" if torch.cuda.is_available() else None,
+        ).eval()
+
+        super(CausalLM, self).__init__(tokenizer=tokenizer, num_heads=self.model.config.num_attention_heads, device=device)
+
+    def forward(
+            self, input_ids, attention_mask, past_key_values: Optional = None
+    ) -> Tuple[torch.Tensor, List[Tuple[torch.Tensor, torch.Tensor]]]:
+        # Model Forward
+        outputs = self.model.forward(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            past_key_values=past_key_values,
+            use_cache=True,
+        )
+        return outputs.logits, outputs.past_key_values

server/text_generation/models/model.py

@@ -1,19 +1,139 @@
+import torch
+
 from abc import ABC, abstractmethod
-from typing import List, Tuple, Optional, TypeVar, Type
+from typing import List, Tuple, Optional
+from tokenizers import Tokenizer
 
 from text_generation.models.types import Batch, GeneratedText
 
-B = TypeVar("B", bound=Batch)
-
 
 class Model(ABC):
-    @property
+    def __init__(self, tokenizer: Tokenizer, num_heads: int, device: torch.device):
-    @abstractmethod
+        self.tokenizer = tokenizer
-    def batch_type(self) -> Type[B]:
+        self.num_heads = num_heads
-        raise NotImplementedError
+        self.device = device
 
     @abstractmethod
-    def generate_token(
+    def forward(self, input_ids, attention_mask, past_key_values: Optional = None) -> Tuple[torch.Tensor, List[Tuple]]:
-            self, batch: B
-    ) -> Tuple[List[GeneratedText], Optional[B]]:
         raise NotImplementedError
+
+    def generate_token(
+            self, batch: Batch
+    ) -> Tuple[List[GeneratedText], Optional[Batch]]:
+        # For some reason, inference_mode does not work well with GLOO which we use on CPU
+        context_manager = (
+            torch.no_grad if self.device.type == "cpu" else torch.inference_mode
+        )
+        with context_manager():
+            logits, past = self.forward(**batch.input_ids)
+
+        # List of indices to cache
+        next_batch_keep_indices = []
+
+        # New input_ids for next forward
+        next_batch_input_ids = []
+        next_batch_all_input_ids = []
+        next_all_input_lengths = []
+
+        next_batch_size = 0
+        next_batch_max_sequence_length = 0
+
+        # Finished requests
+        generated_texts: List[GeneratedText] = []
+
+        # Zipped iterator
+        iterator = zip(
+            batch.requests,
+            batch.all_input_lengths,
+            logits,
+            batch.next_token_choosers,
+            batch.stopping_criterias,
+            batch.all_input_ids,
+        )
+
+        # For each member of the batch
+        for i, (
+                request,
+                input_length,
+                logits,
+                next_token_chooser,
+                stopping_criteria,
+                all_tokens,
+        ) in enumerate(iterator):
+            # Select next token
+            next_token = next_token_chooser(all_tokens, logits.unsqueeze(0)[:, -1])
+
+            # Append next token to all tokens
+            all_tokens = torch.cat([all_tokens, next_token])
+
+            # Evaluate stopping criteria
+            if stopping_criteria(all_tokens):
+                # Decode all tokens
+                output = self.tokenizer.decode(
+                    all_tokens.squeeze(-1), skip_special_tokens=True
+                )
+                # Add to the list of finished generations with the original request
+                generated_texts.append(GeneratedText(request, output, stopping_criteria.current_tokens))
+            # add to the next batch
+            else:
+                next_batch_keep_indices.append(i)
+                next_batch_input_ids.append(next_token)
+                next_batch_all_input_ids.append(all_tokens)
+                next_batch_size += 1
+                new_input_length = input_length + 1
+                next_all_input_lengths.append(new_input_length)
+                next_batch_max_sequence_length = max(
+                    next_batch_max_sequence_length, new_input_length
+                )
+
+        # We finished all generations in the batch; there is no next batch
+        if not next_batch_keep_indices:
+            return generated_texts, None
+
+        # If we finished at least one generation
+        next_batch_input_ids = {"input_ids": torch.cat(next_batch_input_ids, dim=0)}
+        if generated_texts:
+            # Apply indices to attention mask, past key values and other items that need to be cached
+            next_batch_input_ids["attention_mask"] = batch.input_ids["attention_mask"][
+                next_batch_keep_indices
+            ]
+            # Force past to be of dim [batch_size, num_heads, ...] for easy indexing
+            next_batch_input_ids["past_key_values"] = [
+                [t.view(-1, self.num_heads, *t.shape[-2:])[next_batch_keep_indices] for t in layer]
+                for layer in past
+            ]
+            next_batch_requests = [batch.requests[i] for i in next_batch_keep_indices]
+            next_batch_next_token_choosers = [
+                batch.next_token_choosers[i] for i in next_batch_keep_indices
+            ]
+            next_batch_stopping_criterias = [
+                batch.stopping_criterias[i] for i in next_batch_keep_indices
+            ]
+        else:
+            next_batch_input_ids["attention_mask"] = batch.input_ids["attention_mask"]
+            next_batch_input_ids["past_key_values"] = past
+            next_batch_requests = batch.requests
+            next_batch_next_token_choosers = batch.next_token_choosers
+            next_batch_stopping_criterias = batch.stopping_criterias
+
+        # Update attention_mask with padding as we added a new token to input_ids
+        next_batch_input_ids["attention_mask"] = torch.cat(
+            [
+                next_batch_input_ids["attention_mask"],
+                torch.ones((next_batch_size, 1)).to(self.device),
+            ],
+            dim=1,
+        )
+
+        next_batch = Batch(
+            batch_id=batch.batch_id,
+            requests=next_batch_requests,
+            all_input_lengths=next_all_input_lengths,
+            input_ids=next_batch_input_ids,
+            all_input_ids=next_batch_all_input_ids,
+            next_token_choosers=next_batch_next_token_choosers,
+            stopping_criterias=next_batch_stopping_criterias,
+            size=next_batch_size,
+            max_sequence_length=next_batch_max_sequence_length,
+        )
+        return generated_texts, next_batch

server/text_generation/models/types.py

@@ -1,6 +1,5 @@
 import torch
 
-from abc import abstractmethod
 from dataclasses import dataclass
 from typing import List, Dict
 
@@ -51,7 +50,11 @@ class Batch:
                     do_sample=r.parameters.do_sample,
                 )
             )
-            stopping_criterias.append(StoppingCriteria(max_new_tokens=r.max_new_tokens))
+            stopping_criterias.append(
+                StoppingCriteria(
+                    eos_token_id=tokenizer.eos_token_id, max_new_tokens=r.max_new_tokens
+                )
+            )
 
         input_ids = tokenizer(
             inputs, return_tensors="pt", padding=True, pad_to_multiple_of=8
@@ -71,15 +74,171 @@ class Batch:
         )
 
     @classmethod
-    @abstractmethod
     def concatenate(cls, batches: List["Batch"]) -> "Batch":
-        raise NotImplementedError
+        # Used for padding
+        total_batch_size = sum(batch.size for batch in batches)
+        max_sequence_length = max(batch.max_sequence_length for batch in batches)
+        # Only needed for Seq2SeqLM
+        max_encoded_sequence_length = None
+
+        # Batch attributes
+        input_ids = {"input_ids": None, "attention_mask": None, "past_key_values": []}
+        requests = []
+        all_input_lengths = []
+        all_input_ids = []
+        next_token_choosers = []
+        stopping_criterias = []
+
+        # Used for slicing correctly inside the tensors
+        # Equivalent to a cumsum on batch sizes
+        start_index = 0
+        for i, batch in enumerate(batches):
+            requests.extend(batch.requests)
+            all_input_lengths.extend(batch.all_input_lengths)
+            all_input_ids.extend(batch.all_input_ids)
+            next_token_choosers.extend(batch.next_token_choosers)
+            stopping_criterias.extend(batch.stopping_criterias)
+
+            # Slicing end index for this batch
+            end_index = start_index + batch.size
+
+            # We only concatenate batches that did at least one step
+            if batch.input_ids["input_ids"].shape[1] > 1:
+                raise ValueError("Batch input_ids should be of shape (batch_size, 1)")
+
+            # Initialize tensors
+            if i == 0:
+                input_ids["input_ids"] = torch.empty(
+                    (total_batch_size, 1),
+                    dtype=batch.input_ids["input_ids"].dtype,
+                    device=batch.input_ids["input_ids"].device,
+                )
+                input_ids["attention_mask"] = torch.zeros(
+                    (total_batch_size, max_sequence_length),
+                    dtype=batch.input_ids["attention_mask"].dtype,
+                    device=batch.input_ids["attention_mask"].device,
+                )
+
+            # input_ids["input_ids"] is always of shape [batch_size, 1]
+            # We do not need to pad it
+            input_ids["input_ids"][start_index:end_index] = batch.input_ids["input_ids"]
+
+            # We need to slice the attention mask to remove padding from previous steps
+            input_ids["attention_mask"][
+            start_index:end_index, -batch.max_sequence_length:
+            ] = batch.input_ids["attention_mask"][:, -batch.max_sequence_length:]
+
+            for j, past in enumerate(batch.input_ids["past_key_values"]):
+                # Shenanigans to get dimensions because BLOOM outputs a past with a different shape
+                # BLOOM: [batch_size * num_heads, ...] vs [batch_size, num_heads, ...]
+                head_dim, padded_sequence_length = past[0].shape[-2:]
+                num_heads = (
+                    past[0]
+                    .view(batch.size, -1, head_dim, padded_sequence_length)
+                    .shape[1]
+                )
+
+                # This will run only once per layer
+                if j == len(input_ids["past_key_values"]):
+                    input_ids["past_key_values"].append([])
+
+                # Decoder past
+                for k, t in enumerate(past[:2]):
+                    # Needed because BLOOM past shapes are not the same for keys and values
+                    # Keys:   [batch_size * num_heads, head_dim, seq_length]
+                    # Values: [batch_size * num_heads, seq_length, head_dim]
+                    head_dim_last = False
+                    if t.shape[-2] == head_dim:
+                        t = t.view(
+                            batch.size, num_heads, head_dim, padded_sequence_length
+                        )
+                        padded_t_shape = (
+                            total_batch_size,
+                            num_heads,
+                            head_dim,
+                            max_sequence_length - 1,
+                        )
+                    elif t.shape[-1] == head_dim:
+                        head_dim_last = True
+                        t = t.view(
+                            batch.size, num_heads, padded_sequence_length, head_dim
+                        )
+                        padded_t_shape = (
+                            total_batch_size,
+                            num_heads,
+                            max_sequence_length - 1,
+                            head_dim,
+                        )
+                    else:
+                        raise ValueError(f"shape {t.shape} is not valid")
+
+                    # Initialize tensors
+                    # This will run only once per layer and per past tensor
+                    if k == len(input_ids["past_key_values"][j]):
+                        input_ids["past_key_values"][j].append(
+                            torch.zeros(padded_t_shape, dtype=t.dtype, device=t.device)
+                        )
+
+                    # We slice the past keys and values to remove the padding from previous batches
+                    if not head_dim_last:
+                        input_ids["past_key_values"][j][k][
+                        start_index:end_index,
+                        :,
+                        :,
+                        -(batch.max_sequence_length - 1):,
+                        ] = t[:, :, :, -(batch.max_sequence_length - 1):]
+                    else:
+                        input_ids["past_key_values"][j][k][
+                        start_index:end_index,
+                        :,
+                        -(batch.max_sequence_length - 1):,
+                        :,
+                        ] = t[:, :, -(batch.max_sequence_length - 1):, :]
+
+                # Seq2SeqLM specific past (encoder past)
+                for k, t in enumerate(past[2:]):
+                    if max_encoded_sequence_length is None:
+                        max_encoded_sequence_length = max(max(batch.all_input_lengths) for batch in batches)
+                    batch_max_encoded_sequence_length = max(batch.all_input_lengths)
+
+                    padded_t_shape = (total_batch_size, num_heads, max_encoded_sequence_length, head_dim)
+
+                    idx = k + 2
+
+                    # Initialize tensors
+                    # This will run only once per layer and per past tensor
+                    if idx == len(input_ids["past_key_values"][j]):
+                        input_ids["past_key_values"][j].append(
+                            torch.zeros(padded_t_shape, dtype=t.dtype, device=t.device)
+                        )
+
+                    input_ids["past_key_values"][j][idx][
+                    start_index:end_index,
+                    :,
+                    -batch_max_encoded_sequence_length:,
+                    :
+                    ] = t[:, :, -batch_max_encoded_sequence_length:, :]
+
+            start_index += batch.size
+
+        return cls(
+            batch_id=batches[0].batch_id,
+            requests=requests,
+            all_input_lengths=all_input_lengths,
+            input_ids=input_ids,
+            all_input_ids=all_input_ids,
+            next_token_choosers=next_token_choosers,
+            stopping_criterias=stopping_criterias,
+            size=total_batch_size,
+            max_sequence_length=max_sequence_length,
+        )
 
 
 @dataclass
 class GeneratedText:
     request: generate_pb2.Request
     output: str
+    tokens: int
 
     def to_pb(self) -> generate_pb2.GeneratedText:
-        return generate_pb2.GeneratedText(request=self.request, output=self.output)
+        return generate_pb2.GeneratedText(request=self.request, output=self.output, tokens=self.tokens)

server/text_generation/server.py

@@ -27,7 +27,7 @@ class TextGenerationService(generate_pb2_grpc.TextGenerationServiceServicer):
         return generate_pb2.ClearCacheResponse()
 
     async def Generate(self, request, context):
-        batch = self.model.batch_type.from_pb(request.batch, self.model.tokenizer, self.model.device)
+        batch = Batch.from_pb(request.batch, self.model.tokenizer, self.model.device)
 
         generated_texts, next_batch = self.model.generate_token(batch)
         self.cache.set(next_batch)
@@ -51,7 +51,7 @@ class TextGenerationService(generate_pb2_grpc.TextGenerationServiceServicer):
             batches.append(batch)
 
         if len(batches) > 1:
-            batch = self.model.batch_type.concatenate(batches)
+            batch = Batch.concatenate(batches)
         else:
             batch = batches[0]
 

server/text_generation/utils.py

@@ -58,7 +58,8 @@ class NextTokenChooser:
 
 
 class StoppingCriteria:
-    def __init__(self, max_new_tokens=20):
+    def __init__(self, eos_token_id, max_new_tokens=20):
+        self.eos_token_id = eos_token_id
         self.max_new_tokens = max_new_tokens
         self.current_tokens = 0
 
@@ -66,6 +67,8 @@ class StoppingCriteria:
         self.current_tokens += 1
         if self.current_tokens >= self.max_new_tokens:
             return True
+        if self.eos_token_id is not None and all_ids[-1] == self.eos_token_id:
+            return True
         return False
 
 
@@ -124,11 +127,18 @@ def download_weights(model_name, extension=".safetensors"):
     filenames = weight_hub_files(model_name, extension)
 
     download_function = partial(
-        hf_hub_download, repo_id=model_name, local_files_only=False
+        hf_hub_download,
+        repo_id=model_name,
+        local_files_only=False,
     )
 
     executor = ThreadPoolExecutor(max_workers=5)
-    futures = [executor.submit(download_function, filename=filename) for filename in filenames]
+    futures = [
-    files = [file for file in tqdm(concurrent.futures.as_completed(futures), total=len(futures))]
+        executor.submit(download_function, filename=filename) for filename in filenames
+    ]
+    files = [
+        file
+        for file in tqdm(concurrent.futures.as_completed(futures), total=len(futures))
+    ]
 
     return files