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": 75660.30532180605,
-      "required_pages_per_second_upper_estimate": 0.006558043506003459
+      "required_token_positions_per_second": 74915.43169944355,
+      "required_pages_per_second_upper_estimate": 0.0064934797482820135
     },
     {
       "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": 9457.538165225757,
-      "required_pages_per_second_upper_estimate": 0.0037727533769051206
+      "required_token_positions_per_second": 9364.428962430444,
+      "required_pages_per_second_upper_estimate": 0.003735610723803432
     },
     {
       "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": 18915.076330451513,
-      "required_pages_per_second_upper_estimate": 0.004375670015076282
+      "required_token_positions_per_second": 18728.857924860888,
+      "required_pages_per_second_upper_estimate": 0.004332591664274895
     },
     {
       "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": 28372.61449567727,
-      "required_pages_per_second_upper_estimate": 0.0017277433631829632
+      "required_token_positions_per_second": 28093.28688729133,
+      "required_pages_per_second_upper_estimate": 0.0017107337773509595
     },
     {
       "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": 28372.61449567727,
-      "required_pages_per_second_upper_estimate": 0.002184202404300971
+      "required_token_positions_per_second": 28093.28688729133,
+      "required_pages_per_second_upper_estimate": 0.0021626989917791126
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28372.61449567727,
-      "required_pages_per_second_upper_estimate": 0.3522929150040015
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28093.28688729133,
+      "required_pages_per_second_upper_estimate": 0.34882460095721635
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.09405923538275829,
-    "producer_p95_seconds": 0.3748152129992377,
-    "gpu_update_seconds": 0.9744607781638288,
-    "producer_batches_per_second": 10.631598225635873,
-    "gpu_batches_per_second": 1.026208568275364,
-    "producer_headroom": 10.360075480078315,
+    "producer_batch_seconds": 0.14498774356997046,
+    "producer_p95_seconds": 0.38286655800038716,
+    "gpu_update_seconds": 0.9841497049071617,
+    "producer_batches_per_second": 6.897134718959222,
+    "gpu_batches_per_second": 1.0161055731803867,
+    "producer_headroom": 6.787813098368658,
     "sustainable": true,
-    "buffer_seconds": 60.41656824615739,
+    "buffer_seconds": 60.03313199933687,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 375900.9025950626,
-    "upload_bytes_per_second": 5499898.3260054095,
+    "generation_bytes_per_second": 446470.69299554866,
+    "upload_bytes_per_second": 5427424.65607054,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.65958317173872,
+    "estimated_full_shards_per_hour": 95.80221741103978,
     "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": 1.168453648002469,
-      "bytes": 10300928,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 4.622547731996747,
       "bytes": 6328963,
@@ -562,6 +556,12 @@
       "bytes": 840789,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.7087262210043264,
+      "bytes": 7732105,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -385,3 +385,4 @@
 {"step": 393500, "loss": 0.5762318316847086, "lr": 0.0003, "grad_norm": 0.15132665752762026, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57597475623, "tokens_per_second": 149733.49687261274, "data_wait_s": 9.641000360716134e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 393600, "loss": 0.5587992034852505, "lr": 0.0003, "grad_norm": 0.14134306857845638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57612020876, "tokens_per_second": 148340.03096399186, "data_wait_s": 1.0613999620545655e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 393700, "loss": 0.5777887590229511, "lr": 0.0003, "grad_norm": 0.1614428000226198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57626484539, "tokens_per_second": 146853.8779125341, "data_wait_s": 1.1125004675704986e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 393800, "loss": 0.5807689875364304, "lr": 0.0003, "grad_norm": 0.16996640827632079, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57641072471, "tokens_per_second": 149420.9259623452, "data_wait_s": 1.029099803417921e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:371f4e2a9bd3147ed7ddb144090f867fea7a93e44fea41718b37c54b95a83998
-size 825312
+oid sha256:05be30703b149ee0c69b58ce3c8b2ad160993f4286e443a5858d344e7c24dc35
+size 827430