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": 77212.51626013826,
-      "required_pages_per_second_upper_estimate": 0.005341741032597206
+      "required_token_positions_per_second": 77171.0704784196,
+      "required_pages_per_second_upper_estimate": 0.005338873717250446
     },
     {
       "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": 9651.564532517283,
-      "required_pages_per_second_upper_estimate": 0.004040908522178807
+      "required_token_positions_per_second": 9646.38380980245,
+      "required_pages_per_second_upper_estimate": 0.004038739461763899
     },
     {
       "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": 19303.129065034565,
-      "required_pages_per_second_upper_estimate": 0.004020332726500604
+      "required_token_positions_per_second": 19292.7676196049,
+      "required_pages_per_second_upper_estimate": 0.004018174710667296
     },
     {
       "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": 28954.69359755185,
-      "required_pages_per_second_upper_estimate": 0.0017357896699160247
+      "required_token_positions_per_second": 28939.151429407346,
+      "required_pages_per_second_upper_estimate": 0.0017348579406672785
     },
     {
       "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": 28954.69359755185,
-      "required_pages_per_second_upper_estimate": 0.002251100068972154
+      "required_token_positions_per_second": 28939.151429407346,
+      "required_pages_per_second_upper_estimate": 0.002249891733761692
     },
     {
       "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": 28954.69359755185,
-      "required_pages_per_second_upper_estimate": 0.3713997203416048
+      "required_token_positions_per_second": 28939.151429407346,
+      "required_pages_per_second_upper_estimate": 0.37120036209652707
     }
   ],
   "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": 0.33992452188275024,
-    "producer_p95_seconds": 0.28617980199851445,
-    "gpu_update_seconds": 0.9548710956602099,
-    "producer_batches_per_second": 2.9418295404558332,
-    "gpu_batches_per_second": 1.0472617765318233,
-    "producer_headroom": 2.8090679965406333,
+    "producer_batch_seconds": 0.4317469791952533,
+    "producer_p95_seconds": 0.31288839699845994,
+    "gpu_update_seconds": 0.9553839222771644,
+    "producer_batches_per_second": 2.316171387843712,
+    "gpu_batches_per_second": 1.0466996321400226,
+    "producer_headroom": 2.212832905184269,
     "sustainable": true,
-    "buffer_seconds": 60.156879026593224,
+    "buffer_seconds": 60.18918710346136,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 360636.5325174012,
-    "upload_bytes_per_second": 6062548.775496704,
+    "generation_bytes_per_second": 327640.0616291165,
+    "upload_bytes_per_second": 6115953.912940152,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.3842046894219,
+    "estimated_full_shards_per_hour": 70.30393015532609,
     "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": 0.9480016679990513,
-      "bytes": 5853547,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 0.9646089939997182,
       "bytes": 7151890,
@@ -562,6 +556,12 @@
       "bytes": 2120655,
       "pages": 20,
       "shards": 2
+    },
+    {
+      "seconds": 1.0825822449987754,
+      "bytes": 10431134,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -245,3 +245,4 @@
 {"step": 509500, "loss": 0.735479474067688, "lr": 0.0003, "grad_norm": 0.21468405559765869, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74573229358, "tokens_per_second": 153840.5896974438, "data_wait_s": 1.0335003025829792e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 509600, "loss": 0.6320120692253113, "lr": 0.0003, "grad_norm": 0.1685832064861581, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74587943237, "tokens_per_second": 152510.95287292535, "data_wait_s": 1.6240999684669077e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 509700, "loss": 0.6125875115394592, "lr": 0.0003, "grad_norm": 0.17960290333110626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74602682646, "tokens_per_second": 141249.430933694, "data_wait_s": 1.763299951562658e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 509800, "loss": 0.5527863502502441, "lr": 0.0003, "grad_norm": 0.1797644494273301, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74617502095, "tokens_per_second": 153857.28984165288, "data_wait_s": 2.068699905066751e-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:e9db51891329143fc7417a5bb3591e95f0b32dc5372b500a8f72cba063d92f3c
-size 526443
+oid sha256:63772c31db07095e9e121644303c83b86c433fa9c66c14bc627bf5321d522ebf
+size 528561