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": 74081.41132388846,
-      "required_pages_per_second_upper_estimate": 0.006421188975934215
+      "required_token_positions_per_second": 75641.7994566505,
+      "required_pages_per_second_upper_estimate": 0.0065564394645144615
     },
     {
       "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": 9260.176415486058,
-      "required_pages_per_second_upper_estimate": 0.003694022824112836
+      "required_token_positions_per_second": 9455.224932081312,
+      "required_pages_per_second_upper_estimate": 0.003771830593617884
     },
     {
       "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": 18520.352830972115,
-      "required_pages_per_second_upper_estimate": 0.004284357680368048
+      "required_token_positions_per_second": 18910.449864162623,
+      "required_pages_per_second_upper_estimate": 0.004374599763523374
     },
     {
       "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": 27780.529246458176,
-      "required_pages_per_second_upper_estimate": 0.0016916884779367455
+      "required_token_positions_per_second": 28365.674796243937,
+      "required_pages_per_second_upper_estimate": 0.0017273207718972623
     },
     {
       "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": 27780.529246458176,
-      "required_pages_per_second_upper_estimate": 0.002138622042819215
+      "required_token_positions_per_second": 28365.674796243937,
+      "required_pages_per_second_upper_estimate": 0.0021836681670282755
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27780.529246458176,
-      "required_pages_per_second_upper_estimate": 0.3449411977905581
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28365.674796243937,
+      "required_pages_per_second_upper_estimate": 0.35220674716271944
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3227009694217031,
-    "producer_p95_seconds": 0.42656380900007207,
-    "gpu_update_seconds": 0.9952294196671911,
-    "producer_batches_per_second": 3.098844115008554,
-    "gpu_batches_per_second": 1.0047934478609004,
-    "producer_headroom": 3.0840608302190535,
+    "producer_batch_seconds": 0.1585351441327134,
+    "producer_p95_seconds": 0.4597514620036236,
+    "gpu_update_seconds": 0.9746991812675309,
+    "producer_batches_per_second": 6.30774965053097,
+    "gpu_batches_per_second": 1.0259575664150729,
+    "producer_headroom": 6.1481584200130905,
     "sustainable": true,
-    "buffer_seconds": 60.70899459969866,
+    "buffer_seconds": 60.43134923858691,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 402678.20668165607,
-    "upload_bytes_per_second": 4689764.353647351,
+    "generation_bytes_per_second": 482907.27145831025,
+    "upload_bytes_per_second": 4725540.998060371,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.40536928498517,
+    "estimated_full_shards_per_hour": 103.62065894901257,
     "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": 0.987766936996195,
-      "bytes": 2378061,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.0273188179999124,
       "bytes": 9379394,
@@ -562,6 +556,12 @@
       "bytes": 11782468,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.3326379119971534,
+      "bytes": 7233527,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -408,3 +408,4 @@
 {"step": 195800, "loss": 0.6248058266937733, "lr": 0.0003, "grad_norm": 0.15313246632839422, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28657488126, "tokens_per_second": 151671.82703529843, "data_wait_s": 6.287998985499144e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 195900, "loss": 0.596268106251955, "lr": 0.0003, "grad_norm": 0.13300833685127866, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28672190723, "tokens_per_second": 148810.555096693, "data_wait_s": 7.950002327561378e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 196000, "loss": 0.5848307311534882, "lr": 0.0003, "grad_norm": 0.17119020365577028, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28686768471, "tokens_per_second": 148656.61338342412, "data_wait_s": 1.133199839387089e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 196100, "loss": 0.5884687602519989, "lr": 0.0003, "grad_norm": 0.13519273574053875, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28701200508, "tokens_per_second": 146497.09774357386, "data_wait_s": 9.642004442866892e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fa164ce564511f818a3ab4b2c34bf876dd189332a8c3dd8a425e622438959026
-size 874809
+oid sha256:d6db82f0a6312f6c4a62f3fc06586ac086bec52b54532361097a10a5b3b716d0
+size 876927