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": 77199.79045920337,
-      "required_pages_per_second_upper_estimate": 0.005340860632160604
+      "required_token_positions_per_second": 77128.42121932778,
+      "required_pages_per_second_upper_estimate": 0.005335923142546558
     },
     {
       "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": 9649.973807400422,
-      "required_pages_per_second_upper_estimate": 0.004040242518790495
+      "required_token_positions_per_second": 9641.052652415972,
+      "required_pages_per_second_upper_estimate": 0.004036507417493407
     },
     {
       "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": 19299.947614800843,
-      "required_pages_per_second_upper_estimate": 0.004019670114317405
+      "required_token_positions_per_second": 19282.105304831945,
+      "required_pages_per_second_upper_estimate": 0.004015954031684239
     },
     {
       "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": 28949.921422201263,
-      "required_pages_per_second_upper_estimate": 0.001735503585290447
+      "required_token_positions_per_second": 28923.157957247917,
+      "required_pages_per_second_upper_estimate": 0.0017338991564319387
     },
     {
       "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": 28949.921422201263,
-      "required_pages_per_second_upper_estimate": 0.0022507290533292272
+      "required_token_positions_per_second": 28923.157957247917,
+      "required_pages_per_second_upper_estimate": 0.0022486483116491455
     },
     {
       "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": 28949.921422201263,
-      "required_pages_per_second_upper_estimate": 0.37133850800017015
+      "required_token_positions_per_second": 28923.157957247917,
+      "required_pages_per_second_upper_estimate": 0.3709952150081184
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.06754770681268951,
+    "producer_batch_seconds": 0.08520105760143792,
     "producer_p95_seconds": 0.2423732099996414,
-    "gpu_update_seconds": 0.9550284989304724,
-    "producer_batches_per_second": 14.804351578847974,
-    "gpu_batches_per_second": 1.0470891718099415,
-    "producer_headroom": 14.13857766598615,
+    "gpu_update_seconds": 0.9559122154249975,
+    "producer_batches_per_second": 11.736943509291875,
+    "gpu_batches_per_second": 1.0461211645416637,
+    "producer_headroom": 11.21948767228524,
     "sustainable": true,
-    "buffer_seconds": 60.16679543261976,
+    "buffer_seconds": 60.222469571774845,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 311738.8800399147,
-    "upload_bytes_per_second": 5557851.311634484,
+    "generation_bytes_per_second": 311485.74906819547,
+    "upload_bytes_per_second": 5594458.544097143,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.89190674684602,
+    "estimated_full_shards_per_hour": 66.83759073290251,
     "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": 1.0698365700009163,
-      "bytes": 2697855,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 0.9394886710015271,
       "bytes": 6144198,
@@ -562,6 +556,12 @@
       "bytes": 12517532,
       "pages": 41,
       "shards": 4
+    },
+    {
+      "seconds": 1.182365384996956,
+      "bytes": 6177943,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -187,3 +187,4 @@
 {"step": 503700, "loss": 0.5739145278930664, "lr": 0.0003, "grad_norm": 0.14823518795035068, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73724708395, "tokens_per_second": 116951.72401048958, "data_wait_s": 1.0966999980155379e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 503800, "loss": 0.570820689201355, "lr": 0.0003, "grad_norm": 0.15223133617051432, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73739397363, "tokens_per_second": 153187.04565472968, "data_wait_s": 1.0036001185653731e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 503900, "loss": 0.6091482639312744, "lr": 0.0003, "grad_norm": 0.15232177667237645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73754158701, "tokens_per_second": 153796.19281024934, "data_wait_s": 9.94700167211704e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 504000, "loss": 0.5923633575439453, "lr": 0.0003, "grad_norm": 0.16738223139786748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73768891949, "tokens_per_second": 153498.19427815807, "data_wait_s": 1.2784999853465706e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5a7e6117525ef83ab40fb791fe0b7064f7df39a0dff13c42258f16cf5787d86c
-size 402816
+oid sha256:0b20b6bd89e35788e2825ba93525c0893ef829200c16e24916368570e312842c
+size 404934