运行测试脚本 tools/spec_benchmark.py 时出现
self.rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type]
~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^
KeyError: 'default'
return super().getattribute(key)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AttributeError: 'Qwen3Config' object has no attribute 'rope_theta'
if all(common_case_regex.match(k) for k in tied_mapping.keys() | tied_mapping.values()):
^^^^^^^^^^^^^^^^^
AttributeError: 'list' object has no attribute 'keys'
scaling_type = self.config.rope_scaling["type"]
~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^
KeyError: 'type'
等问题,运行结果也特别差
=== Speculative Decoding MT_BENCH Benchmark Results ===
Average Acceptance Length: 1.00
Speedup Ratio: 0.49x
模型构建以及权重读取存在问题
运行测试脚本 tools/spec_benchmark.py 时出现
self.rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type]
~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^
KeyError: 'default'
return super().getattribute(key)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AttributeError: 'Qwen3Config' object has no attribute 'rope_theta'
if all(common_case_regex.match(k) for k in tied_mapping.keys() | tied_mapping.values()):
^^^^^^^^^^^^^^^^^
AttributeError: 'list' object has no attribute 'keys'
scaling_type = self.config.rope_scaling["type"]
~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^
KeyError: 'type'
等问题,运行结果也特别差
=== Speculative Decoding MT_BENCH Benchmark Results ===
Average Acceptance Length: 1.00
Speedup Ratio: 0.49x
模型构建以及权重读取存在问题