MCPcopy Create free account
hub / github.com/NVIDIA/TensorRT-LLM / forward

Method forward

tensorrt_llm/models/commandr/model.py:114–148  ·  view source on GitHub ↗
(
        self,
        input_ids=None,
        position_ids=None,
        use_cache=False,
        attention_mask=None,
        spec_decoding_params=None,
        kv_cache_params=None,
        attention_params=None,
        hidden_states=None,
    )

Source from the content-addressed store, hash-verified

112 dtype=config.dtype)
113
114 def forward(
115 self,
116 input_ids=None,
117 position_ids=None,
118 use_cache=False,
119 attention_mask=None,
120 spec_decoding_params=None,
121 kv_cache_params=None,
122 attention_params=None,
123 hidden_states=None,
124 ):
125 if self.mapping.is_first_pp_rank():
126 hidden_states = self.vocab_embedding(input_ids)
127 else:
128 hidden_states = recv(hidden_states, self.mapping.prev_pp_rank())
129
130 hidden_states = self.layers.forward(
131 hidden_states,
132 use_cache=use_cache,
133 attention_mask=attention_mask,
134 kv_cache_params=kv_cache_params,
135 attention_params=attention_params,
136 spec_decoding_params=spec_decoding_params)
137
138 if use_cache:
139 hidden_states, presents = hidden_states
140
141 if self.mapping.is_last_pp_rank():
142 hidden_states = self.ln_f(hidden_states)
143 else:
144 hidden_states = send(hidden_states, self.mapping.next_pp_rank())
145
146 if use_cache:
147 return (hidden_states, presents)
148 return hidden_states
149
150
151class CohereForCausalLM(DecoderModelForCausalLM):

Callers

nothing calls this directly

Calls 6

recvFunction · 0.85
sendFunction · 0.85
is_first_pp_rankMethod · 0.45
prev_pp_rankMethod · 0.45
is_last_pp_rankMethod · 0.45
next_pp_rankMethod · 0.45

Tested by

no test coverage detected