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": 75407.3266986464,
-      "required_pages_per_second_upper_estimate": 0.006536115960116439
+      "required_token_positions_per_second": 75476.10452085512,
+      "required_pages_per_second_upper_estimate": 0.006542077447429694
     },
     {
       "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": 9425.9158373308,
-      "required_pages_per_second_upper_estimate": 0.0037601387575118875
+      "required_token_positions_per_second": 9434.51306510689,
+      "required_pages_per_second_upper_estimate": 0.00376356832021178
     },
     {
       "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": 18851.8316746616,
-      "required_pages_per_second_upper_estimate": 0.004361039477027212
+      "required_token_positions_per_second": 18869.02613021378,
+      "required_pages_per_second_upper_estimate": 0.004365017111707127
     },
     {
       "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": 28277.7475119924,
-      "required_pages_per_second_upper_estimate": 0.0017219664616051516
+      "required_token_positions_per_second": 28303.53919532067,
+      "required_pages_per_second_upper_estimate": 0.001723537039801339
     },
     {
       "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": 28277.7475119924,
-      "required_pages_per_second_upper_estimate": 0.002176899281288291
+      "required_token_positions_per_second": 28303.53919532067,
+      "required_pages_per_second_upper_estimate": 0.0021788847964668507
     },
     {
       "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": 28277.7475119924,
-      "required_pages_per_second_upper_estimate": 0.35111498456600565
+      "required_token_positions_per_second": 28303.53919532067,
+      "required_pages_per_second_upper_estimate": 0.35143523095373147
     }
   ],
   "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": 0.0937007041719653,
-    "producer_p95_seconds": 0.30818277499929536,
-    "gpu_update_seconds": 0.9777299266242714,
-    "producer_batches_per_second": 10.672278387201215,
-    "gpu_batches_per_second": 1.022777326099262,
-    "producer_headroom": 10.434605964432041,
+    "producer_batch_seconds": 0.07874541979705896,
+    "producer_p95_seconds": 0.27901098299844307,
+    "gpu_update_seconds": 0.9768389673532755,
+    "producer_batches_per_second": 12.699151297652346,
+    "gpu_batches_per_second": 1.0237101850159385,
+    "producer_headroom": 12.405025839861725,
     "sustainable": true,
-    "buffer_seconds": 60.61925545070483,
+    "buffer_seconds": 60.56401597590308,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 364474.32041737513,
-    "upload_bytes_per_second": 5260233.793448558,
+    "generation_bytes_per_second": 364662.6653252356,
+    "upload_bytes_per_second": 5204026.453271514,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.20770463362636,
+    "estimated_full_shards_per_hour": 78.24811906640817,
     "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": 1.2185796390003816,
-      "bytes": 6487750,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.3924951020017033,
       "bytes": 2562340,
@@ -562,6 +556,12 @@
       "bytes": 6135725,
       "pages": 27,
       "shards": 3
+    },
+    {
+      "seconds": 1.0262013009996735,
+      "bytes": 964334,
+      "pages": 29,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -248,3 +248,4 @@
 {"step": 339800, "loss": 0.6158789154142141, "lr": 0.0003, "grad_norm": 0.13946108512276084, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49737582528, "tokens_per_second": 149685.63441450475, "data_wait_s": 1.3250999472802505e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 339900, "loss": 0.5688153691589832, "lr": 0.0003, "grad_norm": 0.15310673850184375, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49752329398, "tokens_per_second": 150021.12086090905, "data_wait_s": 1.387000156682916e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 340000, "loss": 0.561069430783391, "lr": 0.0003, "grad_norm": 0.16010040769861256, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49766917329, "tokens_per_second": 148441.14362928597, "data_wait_s": 1.1295000149402767e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 340100, "loss": 0.589220168069005, "lr": 0.0003, "grad_norm": 0.1648770940891779, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49781441091, "tokens_per_second": 135025.60458267588, "data_wait_s": 1.3221000699559227e-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:c3e09fec27bb95b158e9993471ae707eb2d17911896f63cd7b5541ed267bdbdc
-size 533580
+oid sha256:79ff44345f4eed4b75b294ae0d2b87a04d09c862c97f7d84b89c7af06b749f9a
+size 535698