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": 76463.8529074523,
-      "required_pages_per_second_upper_estimate": 0.005289946764720081
+      "required_token_positions_per_second": 76584.10295515915,
+      "required_pages_per_second_upper_estimate": 0.005298265942038993
     },
     {
       "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": 9557.981613431537,
-      "required_pages_per_second_upper_estimate": 0.004001727308191179
+      "required_token_positions_per_second": 9573.012869394894,
+      "required_pages_per_second_upper_estimate": 0.004008020581174719
     },
     {
       "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": 19115.963226863074,
-      "required_pages_per_second_upper_estimate": 0.003981351018304559
+      "required_token_positions_per_second": 19146.025738789787,
+      "required_pages_per_second_upper_estimate": 0.003987612246737141
     },
     {
       "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": 28673.944840294607,
-      "required_pages_per_second_upper_estimate": 0.001718959210596984
+      "required_token_positions_per_second": 28719.03860818468,
+      "required_pages_per_second_upper_estimate": 0.0017216625131277973
     },
     {
       "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": 28673.944840294607,
-      "required_pages_per_second_upper_estimate": 0.0022292730879786803
+      "required_token_positions_per_second": 28719.03860818468,
+      "required_pages_per_second_upper_estimate": 0.0022327789300856105
     },
     {
       "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": 28673.944840294607,
-      "required_pages_per_second_upper_estimate": 0.36779857672803845
+      "required_token_positions_per_second": 28719.03860818468,
+      "required_pages_per_second_upper_estimate": 0.3683769911646167
     }
   ],
   "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": 0.1978556969451688,
-    "producer_p95_seconds": 0.339821480003593,
-    "gpu_update_seconds": 0.9642203105987398,
-    "producer_batches_per_second": 5.054188559842819,
-    "gpu_batches_per_second": 1.037107379929637,
-    "producer_headroom": 4.87335126299624,
+    "producer_batch_seconds": 0.13702716988234442,
+    "producer_p95_seconds": 0.3561681139981374,
+    "gpu_update_seconds": 0.9627063209602204,
+    "producer_batches_per_second": 7.297822766526007,
+    "gpu_batches_per_second": 1.0387383755853834,
+    "producer_headroom": 7.02566010658199,
     "sustainable": true,
-    "buffer_seconds": 60.74587956772061,
+    "buffer_seconds": 60.65049822049388,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 160,
+    "upload_pages_per_commit": 164,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 632528.0574114125,
-    "upload_bytes_per_second": 5353612.100726624,
+    "generation_bytes_per_second": 634565.0717904804,
+    "upload_bytes_per_second": 5393007.501710405,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.86289831045524,
+    "estimated_full_shards_per_hour": 68.0814462436953,
     "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": 0.8457415819975722,
-      "bytes": 905737,
-      "pages": 27,
-      "shards": 1
-    },
     {
       "seconds": 1.6407027059976826,
       "bytes": 6090920,
@@ -562,6 +556,12 @@
       "bytes": 11993569,
       "pages": 43,
       "shards": 3
+    },
+    {
+      "seconds": 1.4481774710002355,
+      "bytes": 7324064,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -357,3 +357,4 @@
 {"step": 680700, "loss": 0.5364224910736084, "lr": 0.0001, "grad_norm": 0.2100102845041653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99636574961, "tokens_per_second": 151614.97339502323, "data_wait_s": 1.017199974739924e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 680800, "loss": 0.5573974251747131, "lr": 0.0001, "grad_norm": 0.1978600324144116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99651232369, "tokens_per_second": 152168.42304824668, "data_wait_s": 1.774200063664466e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 680900, "loss": 0.5373390913009644, "lr": 0.0001, "grad_norm": 0.19648378488115464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99665809603, "tokens_per_second": 151656.21654411632, "data_wait_s": 1.8298000213690102e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 681000, "loss": 0.5409954786300659, "lr": 0.0001, "grad_norm": 0.19962931514173848, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99680547400, "tokens_per_second": 153265.2447942618, "data_wait_s": 1.2550997780635953e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4dfc65f54dc345c8fa04a58ce61017d5fd4713af734e4129438b7dbcc1df21a4
-size 766008
+oid sha256:806acaecd7c933b4e6fd7cfc336cef093dcf16a50ed8055b04ddbb8d194f926e
+size 768126