update generate_data API calls

Author: npalaska

scripts/test_freeform_skills.py

@@ -49,7 +49,7 @@
 
 ds = Dataset.from_list(samples)
 
-skills_flow = SynthSkillsFlow(client, "mixtral", teacher_model).get_flow()
+skills_flow = SynthSkillsFlow(client, "mixtral", teacher_model, 30).get_flow()
 skills_pipe = Pipeline(skills_flow)
 
 sdg = SDG([skills_pipe])

scripts/test_grounded_skills.py

@@ -97,7 +97,7 @@
 
 ds = Dataset.from_list(samples)
 
-skills_flow = SynthGroundedSkillsFlow(client, "mixtral", teacher_model).get_flow()
+skills_flow = SynthGroundedSkillsFlow(client, "mixtral", teacher_model, 30).get_flow()
 skills_pipe = Pipeline(skills_flow)
 
 sdg = SDG([skills_pipe])

scripts/test_knowledge.py

@@ -38,8 +38,8 @@
 
 ds = Dataset.from_list(samples)
 
-mmlu_flow = MMLUBenchFlow(client, "mixtral", teacher_model).get_flow()
-knowledge_flow = SynthKnowledgeFlow(client, "mixtral", teacher_model).get_flow()
+mmlu_flow = MMLUBenchFlow(client, "mixtral", teacher_model, 30).get_flow()
+knowledge_flow = SynthKnowledgeFlow(client, "mixtral", teacher_model, 30).get_flow()
 knowledge_pipe = Pipeline(knowledge_flow)
 mmlu_pipe = Pipeline(mmlu_flow)
 

src/instructlab/sdg/default_flows.py

@@ -29,10 +29,11 @@ def _get_model_prompt(model_family):
 
 
 class Flow(ABC):
-    def __init__(self, client, model_family, model_id, batched=True) -> None:
+    def __init__(self, client, model_family, model_id, num_instructions_to_generate, batched=True) -> None:
         self.client = client
         self.model_family = model_family
         self.model_id = model_id
+        self.num_instructions_to_generate = num_instructions_to_generate
         self.batched = batched
 
     @abstractmethod
@@ -60,7 +61,7 @@ def get_flow(self) -> list:
                 "gen_kwargs": {
                     "max_tokens": 2048,
                     "temperature": 0.7,
-                    "n": 1
+                    "n": self.num_instructions_to_generate
                 },
                 "drop_duplicates": ["output"],
             }
@@ -280,7 +281,7 @@ def get_flow(self) -> list:
                     "output_cols": ["question"],
                     "batch_kwargs": {
                         "num_procs": 8,
-                        "num_samples": 30,
+                        "num_samples": self.num_instructions_to_generate,
                         "batched": self.batched,
                     },
                 },
@@ -375,16 +376,16 @@ def get_flow(self) -> list:
                     "model_prompt": _get_model_prompt(self.model_family),
                     "output_cols": ["context"],
                     "batch_kwargs": {
-                        "num_samples": 30,
                         "num_procs": 8,
                         "batched": self.batched,
                     }
                 },
                 "gen_kwargs": {
                     "temperature": 0.7,
                     "max_tokens": 2048,
-                    "n": 10
+                    "n": self.num_instructions_to_generate
                 },
+                "drop_duplicates": ["context"],
             },
             {
                 "block_type": LLMBlock,
@@ -396,6 +397,7 @@ def get_flow(self) -> list:
                     "model_prompt": _get_model_prompt(self.model_family),
                     "output_cols": ["question"],
                     "batch_kwargs": {
+                        "num_samples": 3,
                         "num_procs": 8,
                         "batched": self.batched,
                     },
@@ -414,7 +416,6 @@ def get_flow(self) -> list:
                     "batch_kwargs": {
                         "num_procs": 8,
                         "batched": self.batched,
-                        "num_samples": 10,
                     },
                 },
             },

src/instructlab/sdg/generate_data.py

@@ -124,7 +124,7 @@ def _gen_test_data(
             outfile.write("\n")
 
 
-def _sdg_init(pipeline, client, model_family, model_name, num_iters, batched):
+def _sdg_init(pipeline, client, model_family, model_name, num_instructions_to_generate, batched):
     knowledge_flow_types = []
     freeform_skill_flow_types = []
     grounded_skill_flow_types = []
@@ -144,7 +144,7 @@ def _sdg_init(pipeline, client, model_family, model_name, num_iters, batched):
         [
             Pipeline(
                 flow_type(
-                    client, model_family, model_name, num_iters, batched
+                    client, model_family, model_name, num_instructions_to_generate, batched
                 ).get_flow()
             )
             for flow_type in knowledge_flow_types
@@ -154,7 +154,7 @@ def _sdg_init(pipeline, client, model_family, model_name, num_iters, batched):
         [
             Pipeline(
                 flow_type(
-                    client, model_family, model_name, num_iters, batched
+                    client, model_family, model_name, num_instructions_to_generate, batched
                 ).get_flow()
             )
             for flow_type in freeform_skill_flow_types
@@ -164,7 +164,7 @@ def _sdg_init(pipeline, client, model_family, model_name, num_iters, batched):
         [
             Pipeline(
                 flow_type(
-                    client, model_family, model_name, num_iters, batched
+                    client, model_family, model_name, num_instructions_to_generate, batched
                 ).get_flow()
             )
             for flow_type in grounded_skill_flow_types