revert

Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>

Author: syuoni

tensorrt_llm/_torch/pyexecutor/grammar_matcher.py

@@ -1,5 +1,4 @@
 import json
-import os
 from abc import ABC, abstractmethod
 
 import llguidance
@@ -159,11 +158,7 @@ def create(
                 grammar = llguidance.LLMatcher.grammar_from_json_schema(
                     '{"type": "object"}')
             case GuidedDecodingParams.GuideType.JSON_SCHEMA:
-                kwargs = {}
-                if os.environ.get("TRTLLM_XGUIDANCE_LENIENT") == "1":
-                    kwargs = {"defaults": {"lenient": True}}
-                grammar = llguidance.LLMatcher.grammar_from_json_schema(
-                    guide, **kwargs)
+                grammar = llguidance.LLMatcher.grammar_from_json_schema(guide)
             case GuidedDecodingParams.GuideType.REGEX:
                 grammar = llguidance.LLMatcher.grammar_from_regex(guide)
             case GuidedDecodingParams.GuideType.EBNF_GRAMMAR:

tensorrt_llm/evaluate/json_mode_eval.py

@@ -13,6 +13,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import json
+import os
 from typing import Iterable, List, Optional, Union
 
 import click
@@ -56,8 +57,13 @@ def generate_samples(self) -> Iterable[tuple]:
         for i, sample in enumerate(self.data):
             if i >= self.num_samples:
                 break
+            schema = sample["schema"]
+            if os.environ.get("TRTLLM_XGUIDANCE_LENIENT") == "1":
+                schema = json.loads(schema)
+                schema["x-guidance"] = {"lenient": True}
+                schema = json.dumps(schema)
             sampling_args = {
-                "guided_decoding": GuidedDecodingParams(json=sample["schema"])
+                "guided_decoding": GuidedDecodingParams(json=schema)
             }
             yield sample["prompt"], sampling_args, sample["completion"]
 

tensorrt_llm/llmapi/mpi_session.py

@@ -167,8 +167,7 @@ def _start_mpi_pool(self):
         assert not self.mpi_pool, 'MPI session already started'
 
         self.mpi_pool = MPIPoolExecutor(max_workers=self.n_workers,
-                                        path=sys.path,
-                                        env=os.environ)
+                                        path=sys.path)
 
     def __del__(self):
         self.shutdown_abort()