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": 76319.37812626985,
-      "required_pages_per_second_upper_estimate": 0.005279951664130202
+      "required_token_positions_per_second": 76565.79414645793,
+      "required_pages_per_second_upper_estimate": 0.005296999296170754
     },
     {
       "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": 9539.92226578373,
-      "required_pages_per_second_upper_estimate": 0.003994166236453106
+      "required_token_positions_per_second": 9570.724268307242,
+      "required_pages_per_second_upper_estimate": 0.0040070623916907445
     },
     {
       "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": 19079.84453156746,
-      "required_pages_per_second_upper_estimate": 0.003973828446588532
+      "required_token_positions_per_second": 19141.448536614484,
+      "required_pages_per_second_upper_estimate": 0.003986658936232964
     },
     {
       "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": 28619.766797351193,
-      "required_pages_per_second_upper_estimate": 0.001715711319647608
+      "required_token_positions_per_second": 28712.172804921724,
+      "required_pages_per_second_upper_estimate": 0.0017212509187056575
     },
     {
       "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": 28619.766797351193,
-      "required_pages_per_second_upper_estimate": 0.002225060983443856
+      "required_token_positions_per_second": 28712.172804921724,
+      "required_pages_per_second_upper_estimate": 0.0022322451440813925
     },
     {
       "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": 28619.766797351193,
-      "required_pages_per_second_upper_estimate": 0.36710363896501064
+      "required_token_positions_per_second": 28712.172804921724,
+      "required_pages_per_second_upper_estimate": 0.36828892401228464
     }
   ],
   "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": 0.11996299846097713,
-    "producer_p95_seconds": 0.22636888700071722,
-    "gpu_update_seconds": 0.9660456074211922,
-    "producer_batches_per_second": 8.335903677210027,
-    "gpu_batches_per_second": 1.0351478152977138,
-    "producer_headroom": 8.05286313125491,
+    "producer_batch_seconds": 0.067691734976421,
+    "producer_p95_seconds": 0.24887008000223432,
+    "gpu_update_seconds": 0.9629365282748887,
+    "producer_batches_per_second": 14.772852259560626,
+    "gpu_batches_per_second": 1.03849004647431,
+    "producer_headroom": 14.225319067539154,
     "sustainable": true,
-    "buffer_seconds": 60.860873267535105,
+    "buffer_seconds": 60.66500128131799,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 346236.6661931622,
-    "upload_bytes_per_second": 5002557.443374793,
+    "generation_bytes_per_second": 346296.08435846475,
+    "upload_bytes_per_second": 5025204.241395261,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.29432858797216,
+    "estimated_full_shards_per_hour": 74.30707834306199,
     "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": 1.2829982630009908,
-      "bytes": 5992379,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.3964128229999915,
       "bytes": 12688722,
@@ -562,6 +556,12 @@
       "bytes": 965218,
       "pages": 28,
       "shards": 1
+    },
+    {
+      "seconds": 1.2396269109995046,
+      "bytes": 7721145,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -197,3 +197,4 @@
 {"step": 904700, "loss": 0.495917409658432, "lr": 3e-05, "grad_norm": 0.31905249510582967, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132422138280, "tokens_per_second": 153116.43496323537, "data_wait_s": 1.2813001376343891e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 904800, "loss": 0.5306363701820374, "lr": 3e-05, "grad_norm": 0.3199387431222164, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132436887047, "tokens_per_second": 153729.73392596567, "data_wait_s": 1.7258000298170373e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 904900, "loss": 0.5055314302444458, "lr": 3e-05, "grad_norm": 0.3194904783678648, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132451574646, "tokens_per_second": 152710.08105990206, "data_wait_s": 1.3526998372981325e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 905000, "loss": 0.5308002829551697, "lr": 3e-05, "grad_norm": 0.3205761780646895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132466342829, "tokens_per_second": 154075.54590951814, "data_wait_s": 1.4929999451851472e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:159053fd819fce1ea6b1ff4df190e88cdcef847a57177a254080f798136622d6
-size 423996
+oid sha256:6c63b006fe35cbd4e4457c40983d56dd768017ebff8827393dc8d300f0bb5535
+size 426114