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": 76704.64389847236,
-      "required_pages_per_second_upper_estimate": 0.005306605244190924
+      "required_token_positions_per_second": 76577.60915292992,
+      "required_pages_per_second_upper_estimate": 0.005297816685733077
     },
     {
       "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": 9588.080487309046,
-      "required_pages_per_second_upper_estimate": 0.004014329078147733
+      "required_token_positions_per_second": 9572.20114411624,
+      "required_pages_per_second_upper_estimate": 0.004007680728751298
     },
     {
       "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": 19176.16097461809,
-      "required_pages_per_second_upper_estimate": 0.0039938886216405655
+      "required_token_positions_per_second": 19144.40228823248,
+      "required_pages_per_second_upper_estimate": 0.003987274124799324
     },
     {
       "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": 28764.241461927137,
-      "required_pages_per_second_upper_estimate": 0.0017243723551888961
+      "required_token_positions_per_second": 28716.60343234872,
+      "required_pages_per_second_upper_estimate": 0.0017215165280547295
     },
     {
       "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": 28764.241461927137,
-      "required_pages_per_second_upper_estimate": 0.0022362932531377454
+      "required_token_positions_per_second": 28716.60343234872,
+      "required_pages_per_second_upper_estimate": 0.0022325896058755707
     },
     {
       "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": 28764.241461927137,
-      "required_pages_per_second_upper_estimate": 0.36895680483738197
+      "required_token_positions_per_second": 28716.60343234872,
+      "required_pages_per_second_upper_estimate": 0.368345755343681
     }
   ],
   "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": 0.15918086854694025,
-    "producer_p95_seconds": 0.2789580170028785,
-    "gpu_update_seconds": 0.961193433054558,
-    "producer_batches_per_second": 6.2821619779333835,
-    "gpu_batches_per_second": 1.0403733167653044,
-    "producer_headroom": 6.038372838574601,
+    "producer_batch_seconds": 0.1291803320079623,
+    "producer_p95_seconds": 0.2781106210022699,
+    "gpu_update_seconds": 0.9627879587199036,
+    "producer_batches_per_second": 7.741116503233347,
+    "gpu_batches_per_second": 1.0386502977556684,
+    "producer_headroom": 7.453053756360992,
     "sustainable": true,
-    "buffer_seconds": 60.55518628243715,
+    "buffer_seconds": 60.65564139935393,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 330902.0197882963,
-    "upload_bytes_per_second": 5223693.01142689,
+    "generation_bytes_per_second": 331945.11186300864,
+    "upload_bytes_per_second": 5313234.065595549,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.00387044178645,
+    "estimated_full_shards_per_hour": 71.22769371907897,
     "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": 1.2959440570011793,
-      "bytes": 5942943,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.9916904240017175,
       "bytes": 2484673,
@@ -562,6 +556,12 @@
       "bytes": 5933626,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.339776704000542,
+      "bytes": 13443293,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -305,3 +305,4 @@
 {"step": 750500, "loss": 0.5602898001670837, "lr": 5e-05, "grad_norm": 0.2433584197934176, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109851982207, "tokens_per_second": 151543.01189973668, "data_wait_s": 1.3510998542187735e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 750600, "loss": 0.4930821657180786, "lr": 5e-05, "grad_norm": 0.21799323546020904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109866588196, "tokens_per_second": 150365.27921716875, "data_wait_s": 1.4193999959388748e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 750700, "loss": 0.6294246315956116, "lr": 5e-05, "grad_norm": 0.27773659105765264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109881302376, "tokens_per_second": 151934.11129469165, "data_wait_s": 1.4614000974688679e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 750800, "loss": 0.5326976776123047, "lr": 5e-05, "grad_norm": 0.23603058862963358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109896032521, "tokens_per_second": 151875.49988644582, "data_wait_s": 1.4716002624481916e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d4fdf3e0d9d2d6948e10af6fc6be262d4a1c3d13630b5a5451def939ac35f562
-size 655089
+oid sha256:73dab565e5afb0a16e981f4bddf3dc1a1f23084ed1c2b4db95910e6b3bfae0ea
+size 657207