Fix processing of sampling_params within a batched request (#1273)

Introduced a bug in #1214 where when sending a batch request with
`sampling_params` defined, we were improperly attempting to update the
`DecodeConfig` with the `List[SamplingParams]`, instead of updating with
the `SamplingParams` object that corresponded to the specific request
within the batch.

This implements a fix for it.

Author: stbaione

shortfin/python/shortfin_apps/llm/components/generate.py

@@ -9,6 +9,8 @@
 import io
 import json
 import logging
+
+from copy import copy
 from typing import List
 
 import shortfin as sf
@@ -146,7 +148,6 @@ def __init__(
         self.complete_infeed = self.system.create_queue()
 
         self.decode_config = service.server_params.decode_config
-        self.decode_config.update_from_sampling_params(gen_req.sampling_params)
 
     async def run(self):
         logger.debug("Started ClientBatchGenerateProcess: %r", self)
@@ -166,6 +167,12 @@ async def run(self):
             else:
                 input_batch = self.tokenize()
             for index, input_tokens in enumerate(input_batch):
+                decode_config = copy(self.decode_config)
+                decode_config.update_from_sampling_params(
+                    self.gen_req.sampling_params
+                    if self.gen_req.is_single
+                    else self.gen_req.sampling_params[index]
+                )
                 gen_process = GenerateItemProcess(
                     self,
                     self.gen_req,
@@ -175,7 +182,7 @@ async def run(self):
                     else self.gen_req.text[index],
                     input_tokens if is_pretokenized else input_tokens.ids,
                     eos_token_id=self.tokenizer.eos_token_id,
-                    decode_config=self.decode_config,
+                    decode_config=decode_config,
                 )
                 gen_processes.append(gen_process)
                 gen_process.launch()