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": 75157.911387643,
-      "required_pages_per_second_upper_estimate": 0.006514497273095991
+      "required_token_positions_per_second": 75252.7795074329,
+      "required_pages_per_second_upper_estimate": 0.006522720201278328
     },
     {
       "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": 9394.738923455376,
-      "required_pages_per_second_upper_estimate": 0.003747701820430579
+      "required_token_positions_per_second": 9406.597438429113,
+      "required_pages_per_second_upper_estimate": 0.0037524323593542018
     },
     {
       "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": 18789.47784691075,
-      "required_pages_per_second_upper_estimate": 0.0043466150163669415
+      "required_token_positions_per_second": 18813.194876858226,
+      "required_pages_per_second_upper_estimate": 0.004352101533839818
     },
     {
       "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": 28184.21677036613,
-      "required_pages_per_second_upper_estimate": 0.0017162709301579884
+      "required_token_positions_per_second": 28219.79231528734,
+      "required_pages_per_second_upper_estimate": 0.0017184372941932312
     },
     {
       "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": 28184.21677036613,
-      "required_pages_per_second_upper_estimate": 0.002169699025888237
+      "required_token_positions_per_second": 28219.79231528734,
+      "required_pages_per_second_upper_estimate": 0.0021724377298156833
     },
     {
       "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": 28184.21677036613,
-      "required_pages_per_second_upper_estimate": 0.3499536457822632
+      "required_token_positions_per_second": 28219.79231528734,
+      "required_pages_per_second_upper_estimate": 0.3503953749864949
     }
   ],
   "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": 0.22454383727355776,
-    "producer_p95_seconds": 0.2547237070011761,
-    "gpu_update_seconds": 0.9809745725866711,
-    "producer_batches_per_second": 4.453473371356515,
-    "gpu_batches_per_second": 1.0193944144374323,
-    "producer_headroom": 4.368744136992579,
+    "producer_batch_seconds": 0.09057034419524257,
+    "producer_p95_seconds": 0.21676743700299994,
+    "gpu_update_seconds": 0.9797378978236637,
+    "producer_batches_per_second": 11.041141654980345,
+    "gpu_batches_per_second": 1.0206811456628813,
+    "producer_headroom": 10.81742491462373,
     "sustainable": true,
-    "buffer_seconds": 60.82042350037361,
+    "buffer_seconds": 60.74374966506715,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 386103.61538489204,
-    "upload_bytes_per_second": 4750288.743410472,
+    "generation_bytes_per_second": 386194.5324039302,
+    "upload_bytes_per_second": 4789757.688183731,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.84884782943793,
+    "estimated_full_shards_per_hour": 82.86835650528364,
     "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": 0.9439528090006206,
-      "bytes": 7900957,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.7852815879996342,
       "bytes": 838334,
@@ -562,6 +556,12 @@
       "bytes": 5287839,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.096052540000528,
+      "bytes": 12098645,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -157,3 +157,4 @@
 {"step": 252800, "loss": 0.5966698415577412, "lr": 0.0003, "grad_norm": 0.12863293663952916, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37003126292, "tokens_per_second": 151806.60624724728, "data_wait_s": 9.617999239708297e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 252900, "loss": 0.603326765820384, "lr": 0.0003, "grad_norm": 0.16471991318646134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37017867429, "tokens_per_second": 149367.99469737144, "data_wait_s": 1.319799957855139e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 253000, "loss": 0.6166946720331907, "lr": 0.0003, "grad_norm": 0.1520392208692349, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37032638328, "tokens_per_second": 151319.9703398773, "data_wait_s": 1.3380000382312573e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 253100, "loss": 0.5919720446690917, "lr": 0.0003, "grad_norm": 0.1526509503626916, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37047144594, "tokens_per_second": 148602.45958990487, "data_wait_s": 1.5008998161647469e-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:dabda43fe2dc377182af819e50a14b9a928824e9d21c244efcce32af78d69a36
-size 339276
+oid sha256:bc422fced3930ce1b9dcb34db4043c0639f11c440d041b32f6091d8e526a703b
+size 341394