Upload folder using huggingface_hub

app.py

@@ -135,39 +135,9 @@ HF_MODEL_REPOS = [
 ]
 HF_DEFAULT_MODEL_TYPE = os.environ.get("HF_DEFAULT_MODEL_TYPE", "d3pm_cross_attention")
 HF_DEFAULT_INCLUDE_NEG = os.environ.get("HF_DEFAULT_INCLUDE_NEG", "false")
-HF_DEFAULT_TYPE = os.environ.get("HF_DEFAULT_TYPE", "d3pm_encoder_decoder")
-HF_DEFAULT_NEG = os.environ.get("HF_DEFAULT_NEG", "false")
 HF_DEFAULT_NUM_STEPS = os.environ.get("HF_DEFAULT_NUM_STEPS")
 HF_DEFAULT_MODEL_SETTINGS_FILE = os.environ.get("HF_DEFAULT_MODEL_SETTINGS_FILE", "model_settings.json")
 
-# import os
-# # from huggingface_hub import hf_hub_download
-#
-# HF_CHECKPOINT_REPO = os.environ.get("HF_CHECKPOINT_REPO", "bhsinghgrid/devflow2")
-# HF_CHECKPOINT_FILE = os.environ.get("HF_CHECKPOINT_FILE", "best_model.pt")
-
-checkpoint_path = hf_hub_download(
-    repo_id=HF_CHECKPOINT_REPO,
-    filename=HF_CHECKPOINT_FILE,
-    repo_type="model",
-)
-
-
-
-def _download_hf_default_checkpoint():
-    try:
-        cache_dir = Path(".hf_model_cache")
-        cache_dir.mkdir(parents=True, exist_ok=True)
-        ckpt = hf_hub_download(
-            repo_id=HF_DEFAULT_MODEL_REPO,
-            filename=HF_DEFAULT_MODEL_FILE,
-            local_dir=str(cache_dir),
-            local_dir_use_symlinks=False,
-        )
-        return ckpt
-    except Exception:
-        return None
-
 
 def _download_hf_model_settings():
     try:
@@ -177,7 +147,6 @@ def _download_hf_model_settings():
             repo_id=HF_DEFAULT_MODEL_REPO,
             filename=HF_DEFAULT_MODEL_SETTINGS_FILE,
             local_dir=str(cache_dir),
-            local_dir_use_symlinks=False,
         )
         with open(settings_path, "r", encoding="utf-8") as f:
             data = json.load(f)
@@ -203,7 +172,6 @@ def _download_hf_checkpoint(repo_id: str, filename: str = "best_model.pt"):
             repo_id=repo_id,
             filename=filename,
             local_dir=str(cache_dir),
-            local_dir_use_symlinks=False,
         )
     except Exception:
         return None
@@ -216,7 +184,6 @@ def _download_hf_settings_for_repo(repo_id: str):
             repo_id=repo_id,
             filename=HF_DEFAULT_MODEL_SETTINGS_FILE,
             local_dir=str(cache_dir),
-            local_dir_use_symlinks=False,
         )
         with open(settings_path, "r", encoding="utf-8") as f:
             data = json.load(f)
@@ -308,7 +275,11 @@ def default_checkpoint_label():
     if not cps:
         return None
     for item in cps:
-        if item["path"].endswith("ablation_results/T4/best_model.pt"):
+        path = item.get("path")
+        if path and path.endswith("ablation_results/T4/best_model.pt"):
+            return item["label"]
+    for item in cps:
+        if item.get("repo_id") == HF_DEFAULT_MODEL_REPO:
             return item["label"]
     return cps[0]["label"]
 
@@ -458,26 +429,6 @@ def apply_preset(preset_name):
     return presets.get(preset_name, presets["Balanced"])
 
 
-def clean_generated_text(text: str, max_consecutive: int = 2) -> str:
-    text = " ".join(text.split())
-    if not text:
-        return text
-    tokens = text.split()
-    cleaned = []
-    prev = None
-    run = 0
-    for tok in tokens:
-        if tok == prev:
-            run += 1
-        else:
-            prev = tok
-            run = 1
-        if run <= max_consecutive:
-            cleaned.append(tok)
-    out = " ".join(cleaned).replace(" ।", "।").replace(" ॥", "॥")
-    return " ".join(out.split())
-
-
 def save_generation(experiment, record):
     ts = datetime.now().strftime("%Y%m%d")
     path = os.path.join(RESULTS_DIR, f"{experiment}_ui_{ts}.json")
@@ -1027,36 +978,6 @@ def run_single_task(model_bundle, task, output_dir, input_text, task4_phase, tas
     return status, log, task_states, flow
 
 
-def run_all_tasks(model_bundle, output_dir, input_text, task4_phase, task5_cfg, quick_mode):
-    if not model_bundle:
-        raise gr.Error("Load a model first.")
-    logs = []
-    failures = 0
-    used_bundled_any = False
-    for task in ["1", "2", "3", "4", "5"]:
-        if quick_mode:
-            code, log, used_bundled = _run_quick_task(task, model_bundle, input_text, task5_cfg)
-        else:
-            code, log, used_bundled = _run_analysis_cmd(
-                task, model_bundle["ckpt_path"], output_dir, input_text, task4_phase, task5_cfg.get("samples", 50)
-            )
-        logs.append(f"\n\n{'='*22} TASK {task} {'='*22}\n{log}")
-        used_bundled_any = used_bundled_any or used_bundled
-        if code != 0:
-            failures += 1
-    if failures or used_bundled_any:
-        _bundle_task_outputs(model_bundle, output_dir)
-    if failures:
-        logs.append(f"\n\n--- Live input summary ---\n{_live_input_summary(model_bundle, input_text)}")
-    if failures:
-        status = f"Run-all finished with {failures} fallback task(s)."
-    elif used_bundled_any:
-        status = "Run-all loaded from bundled analysis outputs."
-    else:
-        status = "All 5 tasks completed."
-    return status, "".join(logs)
-
-
 def _read_text(path):
     if not os.path.exists(path):
         return "Not found."
@@ -1211,7 +1132,7 @@ CUSTOM_CSS = """
 """
 
 
-with gr.Blocks(title="Sanskrit Diffusion Client Demo", css=CUSTOM_CSS) as demo:
+with gr.Blocks(title="Sanskrit Diffusion Model", css=CUSTOM_CSS) as demo:
     model_state = gr.State(None)
     bg_job_state = gr.State("")
 
@@ -1342,12 +1263,6 @@ with gr.Blocks(title="Sanskrit Diffusion Client Demo", css=CUSTOM_CSS) as demo:
         msg = f"Found {len(choices)} checkpoint(s)." if choices else "No checkpoints found."
         return gr.Dropdown(choices=choices, value=value), msg
 
-    def auto_load_default():
-        choices = list(checkpoint_map().keys())
-        if not choices:
-            return None, "No checkpoints found.", {}, 64, DEFAULT_ANALYSIS_OUT
-        return load_selected_model(default_checkpoint_label())
-
     refresh_btn.click(fn=refresh_checkpoints, outputs=[checkpoint_dropdown, load_status])
     load_btn.click(
         fn=load_selected_model_with_outputs,