linting

include/flexflow/inference.h

@@ -54,7 +54,8 @@ struct GenerationRequest {
                     double emission_time_ms_,
                     bool add_special_tokens_ = true)
       : prompt(prompt_), slo_ratio(slo_ratio_),
-        emission_time_ms(emission_time_ms_), add_special_tokens(add_special_tokens_) {}
+        emission_time_ms(emission_time_ms_),
+        add_special_tokens(add_special_tokens_) {}
 };
 
 struct GenerationResult {

include/flexflow/model.h

@@ -765,45 +765,46 @@ class FFModel {
       bool qk_prod_scaling = true,
       bool position_bias = false,
       char const *name = NULL);
-  Tensor groupquery_self_attention(Tensor const input,
-                                   int embed_dim,
-                                   int num_q_heads,
-                                   int num_kv_heads,
-                                   int kdim = 0,
-                                   int vdim = 0,
-                                   float dropout = 0.0f,
-                                   bool bias = false,
-                                   bool add_bias_kv = false,
-                                   bool add_zero_attn = false,
-                                   DataType data_type = DT_NONE,
-                                   Initializer *kernel_initializer = NULL,
-                                   RotaryEmbeddingMeta rotary_embedding_meta = RotaryEmbeddingMeta(),
-                                   bool scaling_query = false,
-                                   float scaling_factor = 1.0f,
-                                   bool qk_prod_scaling = true,
-                                   bool position_bias = false,
-                                   bool streaming_cache = false,
-                                   char const *name = NULL);
-  Tensor
-      spec_inc_multiquery_self_attention(Tensor const input,
-                                         int embed_dim,
-                                         int num_q_heads,
-                                         int num_kv_heads,
-                                         int kdim = 0,
-                                         int vdim = 0,
-                                         float dropout = 0.0f,
-                                         bool bias = false,
-                                         bool add_bias_kv = false,
-                                         bool add_zero_attn = false,
-                                         DataType data_type = DT_NONE,
-                                         Initializer *kernel_initializer = NULL,
-                                         RotaryEmbeddingMeta rotary_embedding_meta = RotaryEmbeddingMeta(),
-                                         bool scaling_query = false,
-                                         float scaling_factor = 1.0f,
-                                         bool qk_prod_scaling = true,
-                                         bool position_bias = false,
-                                         bool streaming_cache = false,
-                                         char const *name = NULL);
+  Tensor groupquery_self_attention(
+      Tensor const input,
+      int embed_dim,
+      int num_q_heads,
+      int num_kv_heads,
+      int kdim = 0,
+      int vdim = 0,
+      float dropout = 0.0f,
+      bool bias = false,
+      bool add_bias_kv = false,
+      bool add_zero_attn = false,
+      DataType data_type = DT_NONE,
+      Initializer *kernel_initializer = NULL,
+      RotaryEmbeddingMeta rotary_embedding_meta = RotaryEmbeddingMeta(),
+      bool scaling_query = false,
+      float scaling_factor = 1.0f,
+      bool qk_prod_scaling = true,
+      bool position_bias = false,
+      bool streaming_cache = false,
+      char const *name = NULL);
+  Tensor spec_inc_multiquery_self_attention(
+      Tensor const input,
+      int embed_dim,
+      int num_q_heads,
+      int num_kv_heads,
+      int kdim = 0,
+      int vdim = 0,
+      float dropout = 0.0f,
+      bool bias = false,
+      bool add_bias_kv = false,
+      bool add_zero_attn = false,
+      DataType data_type = DT_NONE,
+      Initializer *kernel_initializer = NULL,
+      RotaryEmbeddingMeta rotary_embedding_meta = RotaryEmbeddingMeta(),
+      bool scaling_query = false,
+      float scaling_factor = 1.0f,
+      bool qk_prod_scaling = true,
+      bool position_bias = false,
+      bool streaming_cache = false,
+      char const *name = NULL);
   Tensor inc_multiquery_self_attention_verify(
       Tensor const input,
       int embed_dim,

include/flexflow/ops/spec_inc_multihead_self_attention_params.h

@@ -11,7 +11,8 @@ struct SpecIncMultiHeadSelfAttentionParams {
   LayerID layer_guid;
   int embed_dim, num_q_heads, num_kv_heads, kdim, vdim;
   float dropout, scaling_factor;
-  bool qkv_bias, final_bias, add_zero_attn, scaling_query, qk_prod_scaling, position_bias;
+  bool qkv_bias, final_bias, add_zero_attn, scaling_query, qk_prod_scaling,
+      position_bias;
   RotaryEmbeddingMeta rotary_embedding_meta;
   bool streaming_cache;
   char name[MAX_OPNAME];

inference/incr_decoding/incr_decoding.cc

@@ -398,8 +398,11 @@ void FlexFlow::top_level_task(Task const *task,
         assert(false);
       }
       for (size_t i = 1; i < prompt_json.size(); ++i) {
-        requests.push_back(GenerationRequest(
-            prompt_json[i]["prompt"].get<std::string>(), -1.0, 0, add_special_tokens));
+        requests.push_back(
+            GenerationRequest(prompt_json[i]["prompt"].get<std::string>(),
+                              -1.0,
+                              0,
+                              add_special_tokens));
       }
       PoissonEmissionMachine emission_machine(request_per_second, slo_ratios);
       // ConstantEmissionMachine emission_machine(-1, slo_ratios);
@@ -414,7 +417,8 @@ void FlexFlow::top_level_task(Task const *task,
       std::vector<double> timestamps, ratios;
       for (auto const &json_obj : trace_json) {
         EmissionTrace trace(json_obj);
-        requests.push_back(GenerationRequest(trace.prompt, -1.0, 0, add_special_tokens));
+        requests.push_back(
+            GenerationRequest(trace.prompt, -1.0, 0, add_special_tokens));
         timestamps.push_back(trace.emission_time_ms);
         ratios.push_back(trace.slo_ratio);
       }

inference/models/falcon.cc

@@ -113,11 +113,11 @@ void FALCON::create_falcon_model(FFModel &ff,
             DT_NONE, /*data_type*/
             NULL,    /*kernel_initializer*/
             falcon_config.rotary_embedding_meta,
-            false,   /*scaling query*/
-            1.0f,    /*scaling factor*/
-            true,    /*qk_prod_scaling*/
-            false,   /*position_bias*/
-            false,   /*streaming_cache*/
+            false, /*scaling query*/
+            1.0f,  /*scaling factor*/
+            true,  /*qk_prod_scaling*/
+            false, /*position_bias*/
+            false, /*streaming_cache*/
             std::string("layers_" + std::to_string(i) + "_attention")
                 .c_str() /*name*/
         );
@@ -139,10 +139,10 @@ void FALCON::create_falcon_model(FFModel &ff,
             DT_NONE, /*data_type*/
             nullptr, /*kernel_initializer*/
             falcon_config.rotary_embedding_meta,
-            false,   /*scaling query*/
-            1.0f,    /*scaling factor*/
-            true,    /*qk_prod_scaling*/
-            false,   /*position_bias*/
+            false, /*scaling query*/
+            1.0f,  /*scaling factor*/
+            true,  /*qk_prod_scaling*/
+            false, /*position_bias*/
             std::string("layers_" + std::to_string(i) + "_attention")
                 .c_str() /*name*/
         );
@@ -164,11 +164,11 @@ void FALCON::create_falcon_model(FFModel &ff,
             DT_NONE, /*data_type*/
             nullptr, /*kernel_initializer*/
             falcon_config.rotary_embedding_meta,
-            false,   /*scaling query*/
-            1.0f,    /*scaling factor*/
-            true,    /*qk_prod_scaling*/
-            false,   /*position_bias*/
-            false,   /*streaming_cache*/
+            false, /*scaling query*/
+            1.0f,  /*scaling factor*/
+            true,  /*qk_prod_scaling*/
+            false, /*position_bias*/
+            false, /*streaming_cache*/
             std::string("layers_" + std::to_string(i) + "_attention")
                 .c_str() /*name*/
         );

inference/models/llama.cc

@@ -110,10 +110,10 @@ void LLAMA::create_llama_model(FFModel &ff,
             DT_NONE, /*data_type*/
             NULL,    /*kernel_initializer*/
             llama_config.rotary_embedding_meta,
-            false,   /*scaling query*/
-            1.0f,    /*scaling factor*/
-            true,    /*qk_prod_scaling*/
-            false,   /*position_bias*/
+            false, /*scaling query*/
+            1.0f,  /*scaling factor*/
+            true,  /*qk_prod_scaling*/
+            false, /*position_bias*/
             streaming_cache,
             std::string("layers_" + std::to_string(i) + "_attention")
                 .c_str() /*name*/
@@ -135,10 +135,10 @@ void LLAMA::create_llama_model(FFModel &ff,
             DT_NONE, /*data_type*/
             nullptr, /*kernel_initializer*/
             llama_config.rotary_embedding_meta,
-            false,   /*scaling query*/
-            1.0f,    /*scaling factor*/
-            true,    /*qk_prod_scaling*/
-            false,   /*position_bias*/
+            false, /*scaling query*/
+            1.0f,  /*scaling factor*/
+            true,  /*qk_prod_scaling*/
+            false, /*position_bias*/
             std::string("layers_" + std::to_string(i) + "_attention")
                 .c_str() /*name*/
         );
@@ -152,12 +152,12 @@ void LLAMA::create_llama_model(FFModel &ff,
             llama_config.num_key_value_heads,
             llama_config.hidden_size / llama_config.num_attention_heads,
             llama_config.hidden_size / llama_config.num_attention_heads,
-            0.0f,            /*dropout*/
-            false,           /*qkv_bias*/
-            false,           /*final_bias*/
-            false,           /*add_zero_attn*/
-            DT_NONE,         /*data_type*/
-            nullptr,         /*kernel_initializer*/
+            0.0f,    /*dropout*/
+            false,   /*qkv_bias*/
+            false,   /*final_bias*/
+            false,   /*add_zero_attn*/
+            DT_NONE, /*data_type*/
+            nullptr, /*kernel_initializer*/
             llama_config.rotary_embedding_meta,
             false,           /*scaling query*/
             1.0f,            /*scaling factor*/

inference/models/starcoder.cc

@@ -114,18 +114,18 @@ void STARCODER::create_starcoder_model(
                 startcoder_config.num_attention_heads,
             startcoder_config.hidden_size /
                 startcoder_config.num_attention_heads,
-            startcoder_config.dropout_p, /*dropout*/
-            true,                        /*bias*/
-            false,                       /*add_bias_kv*/
-            false,                       /*add_zero_attn*/
-            DT_NONE,                     /*data_type*/
-            nullptr,                     /*kernel_initializer*/
+            startcoder_config.dropout_p,             /*dropout*/
+            true,                                    /*bias*/
+            false,                                   /*add_bias_kv*/
+            false,                                   /*add_zero_attn*/
+            DT_NONE,                                 /*data_type*/
+            nullptr,                                 /*kernel_initializer*/
             startcoder_config.rotary_embedding_meta, /*apply_rotary_embedding*/
-            false,                       /*scaling query*/
-            1.0f,                        /*scaling factor*/
-            true,                        /*qk_prod_scaling*/
-            false,                       /*position_bias*/
-            false,                       /*streaming_cache*/
+            false,                                   /*scaling query*/
+            1.0f,                                    /*scaling factor*/
+            true,                                    /*qk_prod_scaling*/
+            false,                                   /*position_bias*/
+            false,                                   /*streaming_cache*/
             std::string("layers_" + std::to_string(i) + "_attention")
                 .c_str() /*name*/
         );

inference/spec_infer/spec_infer.cc

@@ -604,8 +604,11 @@ void FlexFlow::top_level_task(Task const *task,
         assert(false);
       }
       for (size_t i = 1; i < prompt_json.size(); ++i) {
-        requests.push_back(GenerationRequest(
-            prompt_json[i]["prompt"].get<std::string>(), -1.0, 0, add_special_tokens));
+        requests.push_back(
+            GenerationRequest(prompt_json[i]["prompt"].get<std::string>(),
+                              -1.0,
+                              0,
+                              add_special_tokens));
       }
       PoissonEmissionMachine emission_machine(request_per_second, slo_ratios);
       //   ConstantEmissionMachine emission_machine(-1, slo_ratios);
@@ -620,7 +623,8 @@ void FlexFlow::top_level_task(Task const *task,
       std::vector<double> timestamps, ratios;
       for (auto const &json_obj : trace_json) {
         EmissionTrace trace(json_obj);
-        requests.push_back(GenerationRequest(trace.prompt, -1.0, 0, add_special_tokens));
+        requests.push_back(
+            GenerationRequest(trace.prompt, -1.0, 0, add_special_tokens));
         timestamps.push_back(trace.emission_time_ms);
         ratios.push_back(trace.slo_ratio);
       }

src/ops/inc_multihead_self_attention.cc

@@ -54,24 +54,25 @@ bool IncMultiHeadSelfAttentionParams::is_valid(
   return is_valid;
 }
 
-Tensor FFModel::inc_multihead_self_attention(const Tensor input,
-                                             int embed_dim,
-                                             int num_heads,
-                                             int kdim,
-                                             int vdim,
-                                             float dropout,
-                                             bool qkv_bias,
-                                             bool final_bias,
-                                             bool add_zero_attn,
-                                             DataType data_type,
-                                             Initializer *kernel_initializer,
-                                             RotaryEmbeddingMeta rotary_embedding_meta,
-                                             bool scaling_query,
-                                             float scaling_factor,
-                                             bool qk_prod_scaling,
-                                             bool position_bias,
-                                             bool streaming_cache,
-                                             char const *name) {
+Tensor FFModel::inc_multihead_self_attention(
+    const Tensor input,
+    int embed_dim,
+    int num_heads,
+    int kdim,
+    int vdim,
+    float dropout,
+    bool qkv_bias,
+    bool final_bias,
+    bool add_zero_attn,
+    DataType data_type,
+    Initializer *kernel_initializer,
+    RotaryEmbeddingMeta rotary_embedding_meta,
+    bool scaling_query,
+    float scaling_factor,
+    bool qk_prod_scaling,
+    bool position_bias,
+    bool streaming_cache,
+    char const *name) {
   return groupquery_self_attention(input,
                                    embed_dim,
                                    num_heads,
@@ -93,25 +94,26 @@ Tensor FFModel::inc_multihead_self_attention(const Tensor input,
                                    name);
 }
 
-Tensor FFModel::groupquery_self_attention(const Tensor input,
-                                          int embed_dim,
-                                          int num_q_heads,
-                                          int num_kv_heads,
-                                          int kdim,
-                                          int vdim,
-                                          float dropout,
-                                          bool qkv_bias,
-                                          bool final_bias,
-                                          bool add_zero_attn,
-                                          DataType data_type,
-                                          Initializer *kernel_initializer,
-                                          RotaryEmbeddingMeta rotary_embedding_meta,
-                                          bool scaling_query,
-                                          float scaling_factor,
-                                          bool qk_prod_scaling,
-                                          bool position_bias,
-                                          bool streaming_cache,
-                                          char const *name) {
+Tensor FFModel::groupquery_self_attention(
+    const Tensor input,
+    int embed_dim,
+    int num_q_heads,
+    int num_kv_heads,
+    int kdim,
+    int vdim,
+    float dropout,
+    bool qkv_bias,
+    bool final_bias,
+    bool add_zero_attn,
+    DataType data_type,
+    Initializer *kernel_initializer,
+    RotaryEmbeddingMeta rotary_embedding_meta,
+    bool scaling_query,
+    float scaling_factor,
+    bool qk_prod_scaling,
+    bool position_bias,
+    bool streaming_cache,
+    char const *name) {
   if (data_type == DT_NONE) {
     data_type = input->data_type;
   }