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": 76823.21372414565,
-      "required_pages_per_second_upper_estimate": 0.005314808179850901
+      "required_token_positions_per_second": 76276.21676505212,
+      "required_pages_per_second_upper_estimate": 0.005276965661013003
     },
     {
       "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": 9602.901715518206,
-      "required_pages_per_second_upper_estimate": 0.004020534416896468
+      "required_token_positions_per_second": 9534.527095631514,
+      "required_pages_per_second_upper_estimate": 0.003991907391374337
     },
     {
       "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": 19205.803431036413,
-      "required_pages_per_second_upper_estimate": 0.004000062363588358
+      "required_token_positions_per_second": 19069.05419126303,
+      "required_pages_per_second_upper_estimate": 0.003971581103263529
     },
     {
       "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": 28808.705146554617,
-      "required_pages_per_second_upper_estimate": 0.0017270378851901983
+      "required_token_positions_per_second": 28603.581286894543,
+      "required_pages_per_second_upper_estimate": 0.0017147410230095755
     },
     {
       "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": 28808.705146554617,
-      "required_pages_per_second_upper_estimate": 0.002239750109042452
+      "required_token_positions_per_second": 28603.581286894543,
+      "required_pages_per_second_upper_estimate": 0.0022238026311983882
     },
     {
       "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": 28808.705146554617,
-      "required_pages_per_second_upper_estimate": 0.36952713724239833
+      "required_token_positions_per_second": 28603.581286894543,
+      "required_pages_per_second_upper_estimate": 0.36689602861551984
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.17638343045308602,
-    "producer_p95_seconds": 0.3411206110031344,
-    "gpu_update_seconds": 0.9597099161295198,
-    "producer_batches_per_second": 5.669466782856212,
-    "gpu_batches_per_second": 1.0419815229511942,
-    "producer_headroom": 5.4410434906740335,
+    "producer_batch_seconds": 0.2407725155467233,
+    "producer_p95_seconds": 0.34118069199757883,
+    "gpu_update_seconds": 0.9665922502042649,
+    "producer_batches_per_second": 4.153297969784031,
+    "gpu_batches_per_second": 1.034562401869739,
+    "producer_headroom": 4.014545630382352,
     "sustainable": true,
-    "buffer_seconds": 60.46172471615975,
+    "buffer_seconds": 60.89531176286869,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 539794.8115283272,
+    "generation_bytes_per_second": 491073.0396221296,
     "upload_bytes_per_second": 4540227.450629848,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.82740077388155,
+    "estimated_full_shards_per_hour": 105.3728427076141,
     "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": [

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

@@ -425,3 +425,4 @@
 {"step": 762500, "loss": 0.5588604211807251, "lr": 5e-05, "grad_norm": 0.2513745749405481, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111607380573, "tokens_per_second": 150719.39320824147, "data_wait_s": 9.732000762596726e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 762600, "loss": 0.5727044939994812, "lr": 5e-05, "grad_norm": 0.2565716037712041, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111621971646, "tokens_per_second": 151604.20425695862, "data_wait_s": 1.292699744226411e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 762700, "loss": 0.5550649762153625, "lr": 5e-05, "grad_norm": 0.27113370539924303, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111636742854, "tokens_per_second": 152260.83182660094, "data_wait_s": 6.713999027851969e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 762800, "loss": 0.5426653623580933, "lr": 5e-05, "grad_norm": 0.23226449571497457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111651462115, "tokens_per_second": 151931.079185711, "data_wait_s": 1.3764998584520072e-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 sha256:373aa4832c11a83bec45e873d48bd67d9e0a042772e68bdc3fa68b04e88ad2bc
-size 910815
+oid sha256:4af30b8d539dc7c9d18cc4375e76a09218472671588e4938562f52a977dcc2a0
+size 912933