notes with sok discussion

Author: Simon Guo

src/loader.py

@@ -73,6 +73,11 @@ def _gpu_context_from_py(py_path: str, gpu_name: str) -> Dict[str, str]:
         "gpu_best_practices_bullets": best_bullets,
     }
 
+# maybe a class-based approach
+# class PromptConstruction:
+#     def __init__(self, prompts_toml: str):
+#         self.cfg = PromptConfig.from_toml(prompts_toml)
+
 def render_prompt_by_option(
     *,
     prompts_toml: str,
@@ -81,6 +86,7 @@ def render_prompt_by_option(
     context: Dict[str, str],
     gpu_specs_py: Optional[str] = None,
     gpu_name: Optional[str] = None,
+    log_prompt: bool=False
 ) -> str:
     """
     New function that uses languages.X and options.Y structure
@@ -143,6 +149,25 @@ def render_prompt_by_option(
             raise ValueError(f"Option '{option}' requires GPU info; provide gpu_specs_py and gpu_name")
         context = {**context, **_gpu_context_from_py(_abs_path(gpu_specs_py), gpu_name)}
 
+
+    # 1. you can either use your our custome ones
+    # by calling one of the options
+    # 2. here if you want to change th eprompt
+
+    # you do it here
+
+
+    # TODO for Sokserey:
+    # 1. base prompt (one-shot examples based on the laguage)
+    # 2. base prompt + hardware info 
+    # 3. base prompt + few-shot examples (specify which one)
+        # load the source -> which one shtye wanna use, compose the paris in here
+
+    # precision 
+
+    # custom prmotp
+    compoonents A, B, C, "my secret prompt "
+
     # Build the prompt from components
     prompt_parts = []
     for component in option_data["components"]:
@@ -167,3 +192,9 @@ def render_prompt_by_option(
         return prompt_text.format(**context).strip() + "\n"
     except KeyError as e:
         raise KeyError(f"Missing placeholder in context: {e.args[0]}. Available: {list(context.keys())}") from e
+
+
+def main():
+    # cal l thsi fie prompt 
+    # call exmmples -> 
+    

src/prompt_constructor_multilang.py

@@ -22,7 +22,7 @@ def get_prompt_for_language(ref_arch_src: str, language: str = "triton", option:
         context={"ref_arch_src": ref_arch_src},
     )
 
-def get_prompt_with_hardware(ref_arch_src: str, language: str, gpu_name: str) -> str:
+def get_prompt_with_hardware_info(ref_arch_src: str, language: str, gpu_name: str) -> str:
     """
     Generate a hardware-aware prompt for a specific language.
     

src/prompts/prompts.toml

@@ -1,26 +1,27 @@
-[meta]
+[metadata]
 version = "1.0"
 default_language = "cuda"
 
 # -------------------------------------------------------------------------
-# Shared Templates: Used by all languages with placeholders
+# Shared Templates: Used by all backends with placeholders
 # -------------------------------------------------------------------------
 [shared]
 problem_statement = """
-You write custom {language_display} to replace the pytorch operators in the given architecture to get speedups.
+You write custom {language_display} to replace the PyTorch operators in the given architecture to get speedups.
 
 You have complete freedom to choose the set of operators you want to replace. You may make the decision to replace some operators with custom {language_display} and leave others unchanged. You may replace multiple operators with custom implementations, consider operator fusion opportunities (combining multiple operators into a single kernel, for example, combining matmul+relu), or algorithmic changes (such as online softmax). You are only limited by your imagination.
 """
 
+#  on {hardware}, with {precision} as inputs and weigths
 instruction = """
-Optimize the architecture named Model with custom {language_display}! Name your optimized output architecture ModelNew. Output the new code in codeblocks. Please generate real code, NOT pseudocode, make sure the code compiles and is fully functional. Just output the new model code, no other text, and NO testing code!
+Optimize the architecture named Model with custom {language_display},! Name your optimized output architecture ModelNew. Output the new code in codeblocks. Please generate real code, NOT pseudocode, make sure the code compiles and is fully functional. Just output the new model code, no other text, and NO testing code!
 """
 
-# Shared example architecture (same for all languages)
-few_shot_example_arch = "src/prompts/model_ex_add.py"
+# Shared example architecture (same refernece for all backends)
+base_one_shot_example_arch = "src/prompts/model_ex_add.py"
 
 # -------------------------------------------------------------------------
-# Languages: Language-specific configuration (minimal, just what varies)
+# Backends: Backend-specific configuration (minimal, just what varies)
 # -------------------------------------------------------------------------
 [languages.triton]
 language_display = "Triton kernels"
@@ -36,6 +37,8 @@ few_shot_new_arch = "src/prompts/model_new_ex_add_cute.py"
 
 # -------------------------------------------------------------------------
 # Options: Different prompt construction modes
+# You should specify...
+# let's sayif you want tomake your ... 
 # -------------------------------------------------------------------------
 [options.basic]
 # Basic prompt: problem statement + architecture + instruction
@@ -55,36 +58,19 @@ components = ["problem_statement", "few_shot_block", "hardware_header", "hardwar
 requires_gpu = true
 requires_example = true
 
-[options.fix_compile]
-# For fixing compilation errors
-description = "Prompt for fixing compilation errors"
-components = ["problem_statement", "arch_with_context", "failed_kernel", "compile_metadata", "fix_footer"]
 
-[options.fix_correctness]
-# For fixing correctness errors
-description = "Prompt for fixing correctness errors"
-components = ["problem_statement", "arch_with_context", "failed_kernel", "correctness_metadata", "fix_footer"]
 
 # -------------------------------------------------------------------------
 # Templates: Reusable text blocks with placeholders
 # -------------------------------------------------------------------------
 [templates.common]
 
-# --- Architecture Presentation ---
+# --- We present the task, a Problem / PyTorch Architecture  ---
 # Used in prompts to present the reference architecture that needs optimization
 arch_block = """
 You are given the following architecture:
 
 
-{ref_arch_src}
-
-"""
-
-# Used in fix prompts to reference the architecture with contextual phrasing
-arch_with_context = """
-With the following architecture:
-
-
 {ref_arch_src}
 
 """
@@ -104,6 +90,9 @@ The example new arch with custom {language_display} looks like this:
 
 """
 
+
+
+
 # --- Error Fix Templates ---
 # Presents a kernel that failed (used in fix_compile and fix_correctness options)
 failed_kernel = """