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": 76109.34579618841,
-      "required_pages_per_second_upper_estimate": 0.006596965196240244
+      "required_token_positions_per_second": 75964.65498686068,
+      "required_pages_per_second_upper_estimate": 0.00658442376360322
     },
     {
       "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": 9513.668224523552,
-      "required_pages_per_second_upper_estimate": 0.003795144496778184
+      "required_token_positions_per_second": 9495.581873357585,
+      "required_pages_per_second_upper_estimate": 0.0037879295808830324
     },
     {
       "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": 19027.336449047103,
-      "required_pages_per_second_upper_estimate": 0.004401639417802764
+      "required_token_positions_per_second": 18991.16374671517,
+      "required_pages_per_second_upper_estimate": 0.004393271499736089
     },
     {
       "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": 28541.004673570656,
-      "required_pages_per_second_upper_estimate": 0.0017379974415416856
+      "required_token_positions_per_second": 28486.745620072754,
+      "required_pages_per_second_upper_estimate": 0.0017346933498589162
     },
     {
       "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": 28541.004673570656,
-      "required_pages_per_second_upper_estimate": 0.00219716554632905
+      "required_token_positions_per_second": 28486.745620072754,
+      "required_pages_per_second_upper_estimate": 0.0021929885341921084
     },
     {
       "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": 28541.004673570656,
-      "required_pages_per_second_upper_estimate": 0.35438375744776507
+      "required_token_positions_per_second": 28486.745620072754,
+      "required_pages_per_second_upper_estimate": 0.3537100415966916
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.10482345510143887,
-    "producer_p95_seconds": 0.34255679300258635,
-    "gpu_update_seconds": 0.9687115193111058,
-    "producer_batches_per_second": 9.539849636059873,
-    "gpu_batches_per_second": 1.0322990695012535,
-    "producer_headroom": 9.24136223494706,
+    "producer_batch_seconds": 0.3486415798597591,
+    "producer_p95_seconds": 0.3502704359998461,
+    "gpu_update_seconds": 0.9705566360138469,
+    "producer_batches_per_second": 2.8682752080295457,
+    "gpu_batches_per_second": 1.0303365748000852,
+    "producer_headroom": 2.783823537067073,
     "sustainable": true,
-    "buffer_seconds": 60.060114197288556,
+    "buffer_seconds": 60.17451143285851,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 500075.920469575,
-    "upload_bytes_per_second": 5152804.8514879225,
+    "generation_bytes_per_second": 442377.06515100045,
+    "upload_bytes_per_second": 5112466.26930223,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.30465136113584,
+    "estimated_full_shards_per_hour": 94.92382017037879,
     "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": 1.585001501996885,
-      "bytes": 10678100,
-      "pages": 41,
-      "shards": 3
-    },
     {
       "seconds": 0.8894862470006046,
       "bytes": 957902,
@@ -562,6 +556,12 @@
       "bytes": 2301578,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 2.074865057998977,
+      "bytes": 9907182,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -361,3 +361,4 @@
 {"step": 191100, "loss": 0.6132036913186312, "lr": 0.0003, "grad_norm": 0.1498338886749033, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27969638045, "tokens_per_second": 152934.77459276258, "data_wait_s": 1.0142000974155962e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 191200, "loss": 0.6133407857269049, "lr": 0.0003, "grad_norm": 0.1524501252896246, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27984303600, "tokens_per_second": 150681.225511353, "data_wait_s": 9.14699921850115e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 191300, "loss": 0.6383128762245178, "lr": 0.0003, "grad_norm": 0.16014402102186312, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27999034189, "tokens_per_second": 151107.54512803836, "data_wait_s": 1.7274993297178298e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 191400, "loss": 0.5785335712134838, "lr": 0.0003, "grad_norm": 0.14933348417125863, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28013607541, "tokens_per_second": 148821.6900789534, "data_wait_s": 1.119099761126563e-05, "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:c8c7654810cadc11477301cd37ae7001f4d3a532f5c4981db882926bf3308c7a
-size 774480
+oid sha256:3a189ed48ced06de8e6dd6a2a6198ca0ecb6a5c9be7041660bdc57ab2446be98
+size 776598