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": 75778.97329776367,
-      "required_pages_per_second_upper_estimate": 0.00656832934539823
+      "required_token_positions_per_second": 75502.79806401806,
+      "required_pages_per_second_upper_estimate": 0.0065443911760968915
     },
     {
       "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": 9472.37166222046,
-      "required_pages_per_second_upper_estimate": 0.0037786706806368516
+      "required_token_positions_per_second": 9437.849758002258,
+      "required_pages_per_second_upper_estimate": 0.0037648993768957468
     },
     {
       "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": 18944.74332444092,
-      "required_pages_per_second_upper_estimate": 0.00438253295201447
+      "required_token_positions_per_second": 18875.699516004515,
+      "required_pages_per_second_upper_estimate": 0.004366560882062235
     },
     {
       "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": 28417.114986661378,
-      "required_pages_per_second_upper_estimate": 0.0017304532096078633
+      "required_token_positions_per_second": 28313.549274006775,
+      "required_pages_per_second_upper_estimate": 0.0017241466010744985
     },
     {
       "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": 28417.114986661378,
-      "required_pages_per_second_upper_estimate": 0.0021876281752821708
+      "required_token_positions_per_second": 28313.549274006775,
+      "required_pages_per_second_upper_estimate": 0.002179655400033778
     },
     {
       "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": 28417.114986661378,
-      "required_pages_per_second_upper_estimate": 0.3528454621684614
+      "required_token_positions_per_second": 28313.549274006775,
+      "required_pages_per_second_upper_estimate": 0.35155952262943463
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.09521060190618869,
-    "producer_p95_seconds": 0.2779873849976866,
-    "gpu_update_seconds": 0.9729347969692775,
-    "producer_batches_per_second": 10.503032015124779,
-    "gpu_batches_per_second": 1.0278181057096851,
-    "producer_headroom": 10.218765321197248,
+    "producer_batch_seconds": 0.4261239306796085,
+    "producer_p95_seconds": 0.3102632200025255,
+    "gpu_update_seconds": 0.9764936120312623,
+    "producer_batches_per_second": 2.3467351350231347,
+    "gpu_batches_per_second": 1.0240722393665644,
+    "producer_headroom": 2.2915718684794126,
     "sustainable": true,
-    "buffer_seconds": 60.32195741209521,
+    "buffer_seconds": 60.54260394593826,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 454373.414685938,
-    "upload_bytes_per_second": 3969190.513530301,
+    "generation_bytes_per_second": 385695.18532543484,
+    "upload_bytes_per_second": 4071480.653287653,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.49795751985172,
+    "estimated_full_shards_per_hour": 82.7612082464436,
     "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": 3.9685035570000764,
-      "bytes": 7473998,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.2958799169973645,
       "bytes": 6217387,
@@ -562,6 +556,12 @@
       "bytes": 7799438,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.8605145479996281,
+      "bytes": 5751347,
+      "pages": 25,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -264,3 +264,4 @@
 {"step": 341400, "loss": 0.559636328369379, "lr": 0.0003, "grad_norm": 0.14600848201033748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49971705475, "tokens_per_second": 148040.5523181139, "data_wait_s": 7.499998901039362e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 341500, "loss": 0.6143929120153189, "lr": 0.0003, "grad_norm": 0.19491709326595896, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49986554279, "tokens_per_second": 151903.6701715982, "data_wait_s": 1.0818999726325274e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 341600, "loss": 0.5268503502011299, "lr": 0.0003, "grad_norm": 0.18204988849407544, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50001276529, "tokens_per_second": 149005.6634609831, "data_wait_s": 1.3790999219054356e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 341700, "loss": 0.6026564370840788, "lr": 0.0003, "grad_norm": 0.14756879543224583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50015922048, "tokens_per_second": 149162.0375101352, "data_wait_s": 1.888599945232272e-05, "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 sha256:02481ef3c644ec1e4b008a3d27f66e52d70c3d93ec372478c694074617917974
-size 567468
+oid sha256:209df16919efd86a62737a26a7b4ca4823998152facb318c500ddf2cc321846e
+size 569586