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": 75398.60078129616,
-      "required_pages_per_second_upper_estimate": 0.006535359619716156
+      "required_token_positions_per_second": 75430.41413839218,
+      "required_pages_per_second_upper_estimate": 0.0065381171208790225
     },
     {
       "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": 9424.82509766202,
-      "required_pages_per_second_upper_estimate": 0.00375970364515
+      "required_token_positions_per_second": 9428.801767299023,
+      "required_pages_per_second_upper_estimate": 0.0037612899981247097
     },
     {
       "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": 18849.65019532404,
-      "required_pages_per_second_upper_estimate": 0.004360534830175193
+      "required_token_positions_per_second": 18857.603534598045,
+      "required_pages_per_second_upper_estimate": 0.004362374695242244
     },
     {
       "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": 28274.475292986062,
-      "required_pages_per_second_upper_estimate": 0.001721767200635672
+      "required_token_positions_per_second": 28286.405301897066,
+      "required_pages_per_second_upper_estimate": 0.0017224936755864328
     },
     {
       "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": 28274.475292986062,
-      "required_pages_per_second_upper_estimate": 0.0021766473768111013
+      "required_token_positions_per_second": 28286.405301897066,
+      "required_pages_per_second_upper_estimate": 0.002177565781921449
     },
     {
       "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": 28274.475292986062,
-      "required_pages_per_second_upper_estimate": 0.3510743545573595
+      "required_token_positions_per_second": 28286.405301897066,
+      "required_pages_per_second_upper_estimate": 0.3512224853408628
     }
   ],
   "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": 0.14670239874985214,
-    "producer_p95_seconds": 0.3197999939984584,
-    "gpu_update_seconds": 0.9778430797921308,
-    "producer_batches_per_second": 6.816521123864772,
-    "gpu_batches_per_second": 1.0226589732706186,
-    "producer_headroom": 6.665488009228046,
+    "producer_batch_seconds": 0.10889561139052262,
+    "producer_p95_seconds": 0.25895845599734457,
+    "gpu_update_seconds": 0.9774306669552582,
+    "producer_batches_per_second": 9.183106529553235,
+    "gpu_batches_per_second": 1.0230904695419947,
+    "producer_headroom": 8.975849939902405,
     "sustainable": true,
-    "buffer_seconds": 60.62627094711211,
+    "buffer_seconds": 60.60070135122601,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 376825.820716839,
-    "upload_bytes_per_second": 5214897.111316019,
+    "generation_bytes_per_second": 375507.4786899974,
+    "upload_bytes_per_second": 5058501.716850791,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.85804906967999,
+    "estimated_full_shards_per_hour": 80.5751635601515,
     "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": 2.1076598479994573,
-      "bytes": 6179293,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.0649322700010089,
       "bytes": 5823579,
@@ -562,6 +556,12 @@
       "bytes": 10536411,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 3.8299359349985025,
+      "bytes": 2330234,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -209,3 +209,4 @@
 {"step": 215900, "loss": 0.5656412690877914, "lr": 0.0003, "grad_norm": 0.1673837270927268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31600938353, "tokens_per_second": 147248.75614039652, "data_wait_s": 1.1638003343250602e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 216000, "loss": 0.6179937608540058, "lr": 0.0003, "grad_norm": 0.1627342998250686, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31615419482, "tokens_per_second": 147161.2405685197, "data_wait_s": 9.397001122124493e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 216100, "loss": 0.5529602710157633, "lr": 0.0003, "grad_norm": 0.13528198204590208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31630058924, "tokens_per_second": 148614.7251212943, "data_wait_s": 1.2836000678362325e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 216200, "loss": 0.5665174443274736, "lr": 0.0003, "grad_norm": 0.16151457878316305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31644759811, "tokens_per_second": 148242.5218466121, "data_wait_s": 6.607002433156595e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:78d332c351a356dfc7efd888f1a034eddbe34c4c4a0c3ee01307abc93dba70b2
-size 450195
+oid sha256:867e608389214d8ea0d9f03a5cff3c038fd73adf685eb8b30baad9aad46571cf
+size 452313