Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77277.96373745862,
-      "required_pages_per_second_upper_estimate": 0.005346268840937297
+      "required_token_positions_per_second": 76603.79859652289,
+      "required_pages_per_second_upper_estimate": 0.0052996285322087205
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9659.745467182327,
-      "required_pages_per_second_upper_estimate": 0.004044333708685719
+      "required_token_positions_per_second": 9575.47482456536,
+      "required_pages_per_second_upper_estimate": 0.004009051350393123
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19319.490934364654,
-      "required_pages_per_second_upper_estimate": 0.004023740472390551
+      "required_token_positions_per_second": 19150.94964913072,
+      "required_pages_per_second_upper_estimate": 0.00398863776740891
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28979.23640154698,
-      "required_pages_per_second_upper_estimate": 0.0017372609735408408
+      "required_token_positions_per_second": 28726.424473696086,
+      "required_pages_per_second_upper_estimate": 0.0017221052844876425
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28979.23640154698,
-      "required_pages_per_second_upper_estimate": 0.0022530081640303895
+      "required_token_positions_per_second": 28726.424473696086,
+      "required_pages_per_second_upper_estimate": 0.0022333531486421333
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28979.23640154698,
-      "required_pages_per_second_upper_estimate": 0.37171452907924446
+      "required_token_positions_per_second": 28726.424473696086,
+      "required_pages_per_second_upper_estimate": 0.3684717291170725
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.04168130542967852,
-    "producer_p95_seconds": 0.12587057199993978,
-    "gpu_update_seconds": 0.9540624058170175,
-    "producer_batches_per_second": 23.99157103385648,
-    "gpu_batches_per_second": 1.0481494647550267,
-    "producer_headroom": 22.889455979890986,
+    "producer_batch_seconds": 0.37665707759374634,
+    "producer_p95_seconds": 0.133691792000036,
+    "gpu_update_seconds": 0.9624587990515991,
+    "producer_batches_per_second": 2.6549348452137065,
+    "gpu_batches_per_second": 1.0390055148182902,
+    "producer_headroom": 2.5552654026846273,
     "sustainable": true,
-    "buffer_seconds": 60.1059315664721,
+    "buffer_seconds": 60.63490434025075,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 64,
+    "upload_pages_per_commit": 60,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 335138.1658228636,
-    "upload_bytes_per_second": 2948488.5039587454,
+    "generation_bytes_per_second": 290863.8755262048,
+    "upload_bytes_per_second": 3848506.5394284977,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.91284876837187,
+    "estimated_full_shards_per_hour": 62.41261672343835,
     "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": [
@@ -220,6 +220,12 @@
       "bytes": 1159359,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 0.9606663819999994,
+      "bytes": 10228939,
+      "pages": 23,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -5,3 +5,4 @@
 {"step": 397600, "loss": 0.5619044899940491, "lr": 0.0003, "grad_norm": 0.16381671188554098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58197534493, "tokens_per_second": 150408.54951693214, "data_wait_s": 1.1349000033078482e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 397700, "loss": 0.5737059116363525, "lr": 0.0003, "grad_norm": 0.18972681856577805, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58212118529, "tokens_per_second": 151286.57820842418, "data_wait_s": 8.487000059176353e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 397800, "loss": 0.5603136420249939, "lr": 0.0003, "grad_norm": 0.1566880131773821, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58226717948, "tokens_per_second": 151606.37351580567, "data_wait_s": 1.219699993271206e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 397900, "loss": 0.575571596622467, "lr": 0.0003, "grad_norm": 0.15858385517622817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58241320545, "tokens_per_second": 150444.64497812663, "data_wait_s": 8.042999979807064e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9fcff4ce7e8bc705668978e085984108d390b4997c0d7cfd025e98dbdde8980d
-size 14991
+oid sha256:510ff187876817eab95f6e3cd9f01b0f206254aaf25d2799e9ef66cf432204e4
+size 17109