Spaces:

build-small-hackathon
/

CommitLens

Paused

App Files Files Community

pkheria commited on 28 days ago

Commit

f8b5831

1 Parent(s): 49fb46a

completed

Browse files

Files changed (3) hide show

.DS_Store +0 -0
app.py +1 -58
index.html +126 -1

.DS_Store CHANGED Viewed

Binary files a/.DS_Store and b/.DS_Store differ

app.py CHANGED Viewed

@@ -155,42 +155,6 @@ def _build_mellum_prompt(user_content: str) -> str:
         add_generation_prompt=True,
     )
-def _generate_batch(prompts: list[str]) -> list[str]:
-    """
-    Single batched model.generate() call for all prompts.
-    Left-padding aligns sequences for parallel decode.
-    max_new_tokens is kept low (256) because SUMMARY_SYSTEM_PROMPT
-    instructs the model to stay under 120 words.
-    """
-    log.info("Batch inference: %d prompts", len(prompts))
-    _tokenizer.padding_side = "left"
-    enc = _tokenizer(
-        prompts,
-        return_tensors="pt",
-        padding=True,
-        truncation=True,
-        max_length=3072,          # cap input so batch fits in VRAM
-    ).to("cuda")
-    log.info("Input shape: %s", enc.input_ids.shape)
-    with torch.no_grad():
-        out = _model.generate(
-            **enc,
-            max_new_tokens=4096,   # ≈ 200 words — enough for our tight prompt
-            use_cache=True,
-            do_sample=False,
-            pad_token_id=_tokenizer.pad_token_id,
-        )
-    results = []
-    for seq in out:
-        new_tok = seq[enc.input_ids.shape[1]:]
-        text = _tokenizer.decode(new_tok, skip_special_tokens=True)
-        results.append(_strip_thinking(text))
-    return results
 def _generate_sequential(prompts: list[str]) -> list[str]:
     """Fallback single-prompt inference when batch would OOM."""
     log.info("Sequential inference: %d prompts", len(prompts))
@@ -202,7 +166,7 @@ def _generate_sequential(prompts: list[str]) -> list[str]:
         with torch.no_grad():
             out = _model.generate(
                 **enc,
-                max_new_tokens=1024,
                 use_cache=True,
                 do_sample=True,
                 temperature=0.4,
@@ -213,27 +177,6 @@ def _generate_sequential(prompts: list[str]) -> list[str]:
         results.append(_strip_thinking(text))
     return results
-def _smart_generate(prompts: list[str]) -> list[str]:
-    """
-    Route to batch or sequential based on estimated token count.
-    Catches OOM and retries sequentially.
-    """
-    estimated_tokens = sum(len(p) for p in prompts) // 4
-    use_sequential = (len(prompts) == 1) or (estimated_tokens > BATCH_TOKEN_BUDGET)
-    if use_sequential:
-        log.info("Routing to sequential (est. %d tokens)", estimated_tokens)
-        return _generate_sequential(prompts)
-    try:
-        return _generate_batch(prompts)
-    except torch.cuda.OutOfMemoryError:
-        log.warning("Batch OOM — retrying sequentially")
-        torch.cuda.empty_cache()
-        return _generate_sequential(prompts)
 # ---------------------------------------------------------------------------
 # Groq final report (pure API call — no GPU needed)
 # ---------------------------------------------------------------------------

         add_generation_prompt=True,
     )
 def _generate_sequential(prompts: list[str]) -> list[str]:
     """Fallback single-prompt inference when batch would OOM."""
     log.info("Sequential inference: %d prompts", len(prompts))
         with torch.no_grad():
             out = _model.generate(
                 **enc,
+                max_new_tokens=256,
                 use_cache=True,
                 do_sample=True,
                 temperature=0.4,
         results.append(_strip_thinking(text))
     return results
 # ---------------------------------------------------------------------------
 # Groq final report (pure API call — no GPU needed)
 # ---------------------------------------------------------------------------

index.html CHANGED Viewed

@@ -1,6 +1,7 @@
 <!DOCTYPE html>
 <html lang="en">
 <head>
 <meta charset="UTF-8"/>
 <meta name="viewport" content="width=device-width, initial-scale=1.0"/>
 <title>CommitLens — AI Code Review</title>
@@ -106,6 +107,64 @@ canvas{position:fixed;inset:0;width:100%;height:100%;pointer-events:none;z-index
 .ov-close{background:none;border:0.5px solid #1a1a2a;border-radius:6px;color:#666;font-size:11px;font-family:'Space Mono',monospace;padding:6px 14px;cursor:pointer;letter-spacing:.1em}
 .ov-close:hover{color:#e6edf3;border-color:#3a3a5a}
 .ov-body{font-size:12px;color:#8a9ab0;line-height:1.9;white-space:pre-wrap;word-break:break-word;max-width:760px}
 </style>
 </head>
 <body>
@@ -376,9 +435,75 @@ function renderFiles(files) {
   `).join('');
 }
 function renderReport(md) {
   const pane = document.getElementById('pane-r');
-  pane.innerHTML = `<pre class="rp">${escHtml(md)}</pre>`;
 }
 function escHtml(s) {

 <!DOCTYPE html>
 <html lang="en">
 <head>
+  <script src="https://cdn.jsdelivr.net/npm/marked/marked.min.js"></script>
 <meta charset="UTF-8"/>
 <meta name="viewport" content="width=device-width, initial-scale=1.0"/>
 <title>CommitLens — AI Code Review</title>
 .ov-close{background:none;border:0.5px solid #1a1a2a;border-radius:6px;color:#666;font-size:11px;font-family:'Space Mono',monospace;padding:6px 14px;cursor:pointer;letter-spacing:.1em}
 .ov-close:hover{color:#e6edf3;border-color:#3a3a5a}
 .ov-body{font-size:12px;color:#8a9ab0;line-height:1.9;white-space:pre-wrap;word-break:break-word;max-width:760px}
+.report-actions{
+  display:flex;
+  justify-content:flex-end;
+  margin-bottom:12px;
+}
+.report-actions button{
+  background:#238636;
+  border:none;
+  color:white;
+  padding:8px 14px;
+  border-radius:6px;
+  cursor:pointer;
+  font-family:'Space Mono',monospace;
+  font-size:10px;
+  letter-spacing:.08em;
+}
+.report-actions button:hover{
+  background:#2ea043;
+}
+.markdown-body{
+  color:#d6dee8;
+  line-height:1.8;
+  font-size:13px;
+}
+.markdown-body h2{
+  color:#58a6ff;
+  margin:20px 0 10px;
+  font-size:18px;
+}
+.markdown-body h3{
+  color:#3fb950;
+  margin:16px 0 8px;
+  font-size:15px;
+}
+.markdown-body ul{
+  padding-left:20px;
+}
+.markdown-body li{
+  margin:6px 0;
+}
+.markdown-body p{
+  margin:10px 0;
+}
+.markdown-body code{
+  background:#0d1117;
+  padding:2px 6px;
+  border-radius:4px;
+}
 </style>
 </head>
 <body>
   `).join('');
 }
+window._currentReport = "";
+function downloadReport() {
+  const html = `
+<!DOCTYPE html>
+<html>
+<head>
+<meta charset="utf-8">
+<title>CommitLens Report</title>
+<style>
+body{
+  max-width:900px;
+  margin:40px auto;
+  padding:20px;
+  font-family:system-ui,sans-serif;
+  line-height:1.7;
+}
+h2{
+  color:#2563eb;
+}
+h3{
+  color:#16a34a;
+}
+code{
+  background:#f3f4f6;
+  padding:2px 6px;
+  border-radius:4px;
+}
+</style>
+</head>
+<body>
+${marked.parse(window._currentReport)}
+</body>
+</html>
+`;
+  const blob = new Blob(
+    [html],
+    { type: "text/html" }
+  );
+  const url = URL.createObjectURL(blob);
+  const a = document.createElement("a");
+  a.href = url;
+  a.download = "commitlens-report.html";
+  a.click();
+  URL.revokeObjectURL(url);
+}
 function renderReport(md) {
+  window._currentReport = md || "";
   const pane = document.getElementById('pane-r');
+  pane.innerHTML = `
+    <div class="report-actions">
+      <button onclick="downloadReport()">
+        ⬇ DOWNLOAD REPORT
+      </button>
+    </div>
+    <div class="rp markdown-body">
+      ${marked.parse(md || "")}
+    </div>
+  `;
 }
 function escHtml(s) {