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": 75582.10195976804,
-      "required_pages_per_second_upper_estimate": 0.006551265036786613
+      "required_token_positions_per_second": 75756.78671713104,
+      "required_pages_per_second_upper_estimate": 0.006566406268820368
     },
     {
       "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": 9447.762744971005,
-      "required_pages_per_second_upper_estimate": 0.0037688538156099433
+      "required_token_positions_per_second": 9469.59833964138,
+      "required_pages_per_second_upper_estimate": 0.003777564360795189
     },
     {
       "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": 18895.52548994201,
-      "required_pages_per_second_upper_estimate": 0.00437114727220745
+      "required_token_positions_per_second": 18939.19667928276,
+      "required_pages_per_second_upper_estimate": 0.004381249833274751
     },
     {
       "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": 28343.288234913012,
-      "required_pages_per_second_upper_estimate": 0.0017259575477654191
+      "required_token_positions_per_second": 28408.79501892414,
+      "required_pages_per_second_upper_estimate": 0.0017299465672241624
     },
     {
       "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": 28343.288234913012,
-      "required_pages_per_second_upper_estimate": 0.002181944787567053
+      "required_token_positions_per_second": 28408.79501892414,
+      "required_pages_per_second_upper_estimate": 0.0021869876811346136
     },
     {
       "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": 28343.288234913012,
-      "required_pages_per_second_upper_estimate": 0.3519287809939905
+      "required_token_positions_per_second": 28408.79501892414,
+      "required_pages_per_second_upper_estimate": 0.3527421560143057
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.27771661210941545,
-    "producer_p95_seconds": 0.20397340700037603,
-    "gpu_update_seconds": 0.9754690341801427,
-    "producer_batches_per_second": 3.600792881651666,
-    "gpu_batches_per_second": 1.0251478672928607,
-    "producer_headroom": 3.5124619545474838,
+    "producer_batch_seconds": 0.07048061157813379,
+    "producer_p95_seconds": 0.1901399059988762,
+    "gpu_update_seconds": 0.9732197364084839,
+    "producer_batches_per_second": 14.188299130909419,
+    "gpu_batches_per_second": 1.0275171809506705,
+    "producer_headroom": 13.808332740268387,
     "sustainable": true,
-    "buffer_seconds": 60.47908011916885,
+    "buffer_seconds": 60.339623657326,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 525020.6564935704,
+    "generation_bytes_per_second": 491235.659371762,
     "upload_bytes_per_second": 5577200.94315118,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 112.65721102814993,
+    "estimated_full_shards_per_hour": 105.40773712029119,
     "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": [

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -79,3 +79,4 @@
 {"step": 322900, "loss": 0.5566022116690874, "lr": 0.0003, "grad_norm": 0.1821681413913706, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47263453964, "tokens_per_second": 149769.78794203443, "data_wait_s": 1.5747000361443497e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 323000, "loss": 0.5394137669354677, "lr": 0.0003, "grad_norm": 0.16496411775301884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47278053916, "tokens_per_second": 148925.1525806816, "data_wait_s": 1.7264999769395217e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 323100, "loss": 0.5890489425510168, "lr": 0.0003, "grad_norm": 0.16163307237326052, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47292538793, "tokens_per_second": 147737.5346600004, "data_wait_s": 1.662700015003793e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 323200, "loss": 0.6068186368793249, "lr": 0.0003, "grad_norm": 0.1654916509013183, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47307095938, "tokens_per_second": 148288.541725589, "data_wait_s": 9.934999980032444e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ec1f9b7c69f7e717ffac90f342d40c66b696cabc0cb010d97513a26538a5a450
-size 172506
+oid sha256:0b76e202bd4abda7539dd0a360ff068e6aaab8eaa7acf991ba3548f5911fe5a1
+size 174624