Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75779.87907209457,
-      "required_pages_per_second_upper_estimate": 0.0065684078556215685
+      "required_token_positions_per_second": 75703.13066892742,
+      "required_pages_per_second_upper_estimate": 0.006561755498552089
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9472.484884011821,
-      "required_pages_per_second_upper_estimate": 0.003778715846502242
+      "required_token_positions_per_second": 9462.891333615928,
+      "required_pages_per_second_upper_estimate": 0.0037748888358129598
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18944.969768023642,
-      "required_pages_per_second_upper_estimate": 0.004382585335752062
+      "required_token_positions_per_second": 18925.782667231855,
+      "required_pages_per_second_upper_estimate": 0.004378146737665323
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28417.454652035463,
-      "required_pages_per_second_upper_estimate": 0.0017304738934470585
+      "required_token_positions_per_second": 28388.67400084778,
+      "required_pages_per_second_upper_estimate": 0.0017287213027901355
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28417.454652035463,
-      "required_pages_per_second_upper_estimate": 0.002187654323662922
+      "required_token_positions_per_second": 28388.67400084778,
+      "required_pages_per_second_upper_estimate": 0.0021854387094645523
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28417.454652035463,
-      "required_pages_per_second_upper_estimate": 0.3528496796756207
+      "required_token_positions_per_second": 28388.67400084778,
+      "required_pages_per_second_upper_estimate": 0.3524923203105129
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11683274607787553,
-    "producer_p95_seconds": 0.3722418299948913,
-    "gpu_update_seconds": 0.9729231677693432,
-    "producer_batches_per_second": 8.55924416373338,
-    "gpu_batches_per_second": 1.0278303910603104,
-    "producer_headroom": 8.327486945490742,
+    "producer_batch_seconds": 0.641841748124989,
+    "producer_p95_seconds": 0.3580939550010953,
+    "gpu_update_seconds": 0.9739095245933057,
+    "producer_batches_per_second": 1.5580164470779565,
+    "gpu_batches_per_second": 1.0267894242204783,
+    "producer_headroom": 1.517367057282244,
     "sustainable": true,
-    "buffer_seconds": 60.321236401699274,
+    "buffer_seconds": 60.38239052478495,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 268201.97333914856,
-    "upload_bytes_per_second": 4864540.9685207205,
+    "generation_bytes_per_second": 238775.77894578734,
+    "upload_bytes_per_second": 5063752.244111188,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.54990005617945,
+    "estimated_full_shards_per_hour": 51.23572374611106,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 2.558325848000095,
-      "bytes": 890978,
-      "pages": 28,
-      "shards": 1
-    },
     {
       "seconds": 2.274885417002224,
       "bytes": 15797221,
@@ -562,6 +556,12 @@
       "bytes": 796711,
       "pages": 25,
       "shards": 1
+    },
+    {
+      "seconds": 1.6776708719989983,
+      "bytes": 13189069,
+      "pages": 31,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -329,3 +329,4 @@
 {"step": 270000, "loss": 0.6172356083989143, "lr": 0.0003, "grad_norm": 0.14430560851862967, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39519825793, "tokens_per_second": 149868.97792367495, "data_wait_s": 1.539799995953217e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 270100, "loss": 0.6445578038692474, "lr": 0.0003, "grad_norm": 0.2480123810749485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39534484321, "tokens_per_second": 128718.24655918863, "data_wait_s": 1.6833000699989498e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 270200, "loss": 0.5957466680556536, "lr": 0.0003, "grad_norm": 0.15771897321231884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39549168780, "tokens_per_second": 149595.2218584798, "data_wait_s": 1.2221003999002278e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 270300, "loss": 0.6106796804815531, "lr": 0.0003, "grad_norm": 0.14322447692362245, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39563684915, "tokens_per_second": 120176.25725778555, "data_wait_s": 1.2378004612401128e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6ca3d242e67d55ef45c0303c0c2fbb52bd28308bd548f295d71da307dca1c2c7
-size 706704
+oid sha256:3d276e5d9ad6f9984f10f1781e64ce5534b2e617222c204666f85f242643bc32
+size 708822