Flashinfer sample kernel (#617)

python/sglang/bench_latency.py

@@ -156,14 +156,14 @@ def extend(reqs, model_runner):
     )
     batch.prepare_for_extend(model_runner.model_config.vocab_size, None)
     output = model_runner.forward(batch, ForwardMode.EXTEND)
-    next_token_ids, _ = batch.sample(output.next_token_logits)
+    next_token_ids = batch.sample(output.next_token_logits)
     return next_token_ids, output.next_token_logits, batch
 
 
 def decode(input_token_ids, batch, model_runner):
     batch.prepare_for_decode(input_token_ids.cpu().numpy())
     output = model_runner.forward(batch, ForwardMode.DECODE)
-    next_token_ids, _ = batch.sample(output.next_token_logits)
+    next_token_ids = batch.sample(output.next_token_logits)
     return next_token_ids, output.next_token_logits
 
 

python/sglang/srt/managers/controller/infer_batch.py

@@ -7,6 +7,7 @@ from typing import List, Union
 
 import numpy as np
 import torch
+from flashinfer.sampling import top_k_top_p_sampling_from_probs
 
 from sglang.srt.constrained import RegexGuide
 from sglang.srt.constrained.jump_forward import JumpForwardMap
@@ -398,10 +399,10 @@ class Batch:
         ).view(-1, 1)
         self.top_ps = torch.tensor(
             [r.sampling_params.top_p for r in reqs], dtype=torch.float, device=device
-        ).view(-1, 1)
+        )
         self.top_ks = torch.tensor(
             [r.sampling_params.top_k for r in reqs], dtype=torch.int, device=device
-        ).view(-1, 1)
+        )
         self.frequency_penalties = torch.tensor(
             [r.sampling_params.frequency_penalty for r in reqs],
             dtype=torch.float,
@@ -659,20 +660,17 @@ class Batch:
 
         # TODO(lmzheng): apply penalty
         probs = torch.softmax(logits, dim=-1)
-        probs_sort, probs_idx = _top_p_top_k(probs, self.top_ps, self.top_ks)
         try:
-            sampled_index = torch.multinomial(probs_sort, num_samples=1)
+            max_top_k_round, batch_size = 32, probs.shape[0]
+            uniform_samples = torch.rand(
+                (max_top_k_round, batch_size), device=probs.device
+            )
+            batch_next_token_ids, _ = top_k_top_p_sampling_from_probs(
+                probs, uniform_samples, self.top_ks, self.top_ps
+            )
         except RuntimeError as e:
             warnings.warn(f"Ignore errors in sampling: {e}")
-            sampled_index = torch.ones(
-                probs_sort.shape[:-1] + (1,), dtype=torch.int64, device=probs.device
-            )
-        batch_next_token_ids = torch.gather(probs_idx, dim=1, index=sampled_index).view(
-            -1
-        )
-        batch_next_token_probs = torch.gather(
-            probs_sort, dim=1, index=sampled_index
-        ).view(-1)
+            batch_next_token_ids = torch.argmax(probs, dim=-1)
 
         if has_regex:
             batch_next_token_ids_cpu = batch_next_token_ids.cpu().numpy()
@@ -682,18 +680,7 @@ class Batch:
                         req.regex_fsm_state, batch_next_token_ids_cpu[i]
                     )
 
-        return batch_next_token_ids, batch_next_token_probs
-
-
-def _top_p_top_k(probs: torch.Tensor, top_ps: torch.Tensor, top_ks: torch.Tensor):
-    probs_sort, probs_idx = probs.sort(dim=-1, descending=True)
-    probs_sum = torch.cumsum(probs_sort, dim=-1)
-    probs_sort[(probs_sum - probs_sort) > top_ps] = 0.0
-    probs_sort[
-        torch.arange(0, probs.shape[-1], device=probs.device).view(1, -1) >= top_ks
-    ] = 0.0
-    probs_sort.div_(probs_sort.max(dim=-1, keepdim=True)[0])
-    return probs_sort, probs_idx
+        return batch_next_token_ids
 
 
 @dataclass

python/sglang/srt/managers/controller/tp_worker.py

@@ -451,7 +451,7 @@ class ModelTpServer:
         # Forward and sample the next tokens
         if batch.extend_num_tokens != 0:
             output = self.model_runner.forward(batch, ForwardMode.EXTEND)
-            next_token_ids, _ = batch.sample(output.next_token_logits)
+            next_token_ids = batch.sample(output.next_token_logits)
 
             # Move logprobs to cpu
             if output.next_token_logprobs is not None:
@@ -574,7 +574,7 @@ class ModelTpServer:
 
         # Forward and sample the next tokens
         output = self.model_runner.forward(batch, ForwardMode.DECODE)
-        next_token_ids, _ = batch.sample(output.next_token_logits)
+        next_token_ids = batch.sample(output.next_token_logits)
 
         # Move logprobs to cpu
         if output.next_token_logprobs is not None:

python/sglang/srt/server.py

@@ -154,7 +154,7 @@ def launch_server(server_args: ServerArgs, pipe_finish_writer, model_overide_arg
     if not server_args.disable_flashinfer:
         assert_pkg_version(
             "flashinfer",
-            "0.0.8",
+            "0.1.0",
             "Please uninstall the old version and "
             "reinstall the latest version by following the instructions "
             "at https://docs.flashinfer.ai/installation.html.",