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": 76693.0175741747,
-      "required_pages_per_second_upper_estimate": 0.005305800908099222
+      "required_token_positions_per_second": 76707.53229403529,
+      "required_pages_per_second_upper_estimate": 0.005306805070097969
     },
     {
       "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": 9586.627196771837,
-      "required_pages_per_second_upper_estimate": 0.004013720615747961
+      "required_token_positions_per_second": 9588.441536754412,
+      "required_pages_per_second_upper_estimate": 0.004014480242010937
     },
     {
       "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": 19173.254393543673,
-      "required_pages_per_second_upper_estimate": 0.003993283257454462
+      "required_token_positions_per_second": 19176.883073508823,
+      "required_pages_per_second_upper_estimate": 0.0039940390157964765
     },
     {
       "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": 28759.88159031551,
-      "required_pages_per_second_upper_estimate": 0.0017241109875429175
+      "required_token_positions_per_second": 28765.324610263233,
+      "required_pages_per_second_upper_estimate": 0.0017244372882777735
     },
     {
       "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": 28759.88159031551,
-      "required_pages_per_second_upper_estimate": 0.0022359542922969896
+      "required_token_positions_per_second": 28765.324610263233,
+      "required_pages_per_second_upper_estimate": 0.0022363774631566106
     },
     {
       "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": 28759.88159031551,
-      "required_pages_per_second_upper_estimate": 0.3689008810856134
+      "required_token_positions_per_second": 28765.324610263233,
+      "required_pages_per_second_upper_estimate": 0.36897069830124335
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09538105792981355,
-    "producer_p95_seconds": 0.28471346399965114,
-    "gpu_update_seconds": 0.961339145753301,
-    "producer_batches_per_second": 10.484261987698366,
-    "gpu_batches_per_second": 1.0402156246497218,
-    "producer_headroom": 10.078931463107754,
+    "producer_batch_seconds": 0.09276190921886496,
+    "producer_p95_seconds": 0.2705246689984051,
+    "gpu_update_seconds": 0.9611572396487199,
+    "producer_batches_per_second": 10.780286956368837,
+    "gpu_batches_per_second": 1.0404124931374144,
+    "producer_headroom": 10.361550853604571,
     "sustainable": true,
-    "buffer_seconds": 60.56436618245797,
+    "buffer_seconds": 60.552906097869354,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 157,
+    "upload_pages_per_commit": 167,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 596399.2213471263,
-    "upload_bytes_per_second": 4000997.738953311,
+    "generation_bytes_per_second": 613626.3157193315,
+    "upload_bytes_per_second": 3954040.5828353614,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.98669471888705,
+    "estimated_full_shards_per_hour": 65.83496143190841,
     "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": 1.719156404997193,
-      "bytes": 11941861,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.5756286089999776,
       "bytes": 6163874,
@@ -562,6 +556,12 @@
       "bytes": 7330991,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.5559647560003214,
+      "bytes": 6263907,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -243,3 +243,4 @@
 {"step": 824300, "loss": 0.5308011174201965, "lr": 3e-05, "grad_norm": 0.28065418250414914, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120654415350, "tokens_per_second": 150776.0271253589, "data_wait_s": 1.1844000255223364e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 824400, "loss": 0.5129418969154358, "lr": 3e-05, "grad_norm": 0.27516108330357675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120669000089, "tokens_per_second": 150402.08358406412, "data_wait_s": 1.4411001757252961e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 824500, "loss": 0.5239776968955994, "lr": 3e-05, "grad_norm": 0.274425546399414, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120683638925, "tokens_per_second": 150804.11986326615, "data_wait_s": 1.3046999811194837e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 824600, "loss": 0.5294896364212036, "lr": 3e-05, "grad_norm": 0.2926560772342064, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120698312369, "tokens_per_second": 151075.42040384788, "data_wait_s": 1.2972999684279785e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9bed6dcf99b0c306e0b257a2072a2248f35af593b9576793c03aa57ee19eeb78
-size 522207
+oid sha256:fb9315188e6f30f523ddf86a07684bd46718e3539779c5bb421212b9d45228cd
+size 524325