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": 77100.78421981164,
-      "required_pages_per_second_upper_estimate": 0.005334011150793358
+      "required_token_positions_per_second": 76331.21995640192,
+      "required_pages_per_second_upper_estimate": 0.00528077090941557
     },
     {
       "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": 9637.598027476455,
-      "required_pages_per_second_upper_estimate": 0.004035061038172003
+      "required_token_positions_per_second": 9541.40249455024,
+      "required_pages_per_second_upper_estimate": 0.003994785977326952
     },
     {
       "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": 19275.19605495291,
-      "required_pages_per_second_upper_estimate": 0.004014515017143608
+      "required_token_positions_per_second": 19082.80498910048,
+      "required_pages_per_second_upper_estimate": 0.0039744450318201455
     },
     {
       "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": 28912.794082429366,
-      "required_pages_per_second_upper_estimate": 0.001733277857961287
+      "required_token_positions_per_second": 28624.20748365072,
+      "required_pages_per_second_upper_estimate": 0.0017159775320107293
     },
     {
       "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": 28912.794082429366,
-      "required_pages_per_second_upper_estimate": 0.002247842565967871
+      "required_token_positions_per_second": 28624.20748365072,
+      "required_pages_per_second_upper_estimate": 0.0022254062272711295
     },
     {
       "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": 28912.794082429366,
-      "required_pages_per_second_upper_estimate": 0.3708622783498078
+      "required_token_positions_per_second": 28624.20748365072,
+      "required_pages_per_second_upper_estimate": 0.3671605993208235
     }
   ],
   "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": 0.06118474864071288,
-    "producer_p95_seconds": 0.20401876299911237,
-    "gpu_update_seconds": 0.9562548649285336,
-    "producer_batches_per_second": 16.343942276729255,
-    "gpu_batches_per_second": 1.0457463137452752,
-    "producer_headroom": 15.628974314233483,
+    "producer_batch_seconds": 0.13515331412499165,
+    "producer_p95_seconds": 0.23005809700043756,
+    "gpu_update_seconds": 0.9658957375777723,
+    "producer_batches_per_second": 7.399004652414118,
+    "gpu_batches_per_second": 1.0353084303982465,
+    "producer_headroom": 7.146667056084903,
     "sustainable": true,
-    "buffer_seconds": 60.24405649049762,
+    "buffer_seconds": 60.85143146739965,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 294226.6005242434,
-    "upload_bytes_per_second": 4218781.812410461,
+    "generation_bytes_per_second": 305420.8732146183,
+    "upload_bytes_per_second": 4166831.241528759,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.134179227785836,
+    "estimated_full_shards_per_hour": 65.53620955780899,
     "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": 1.223384894999981,
-      "bytes": 7577595,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 0.9591294869997,
       "bytes": 6044241,
@@ -562,6 +556,12 @@
       "bytes": 6338461,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.8834309400008351,
+      "bytes": 840516,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -135,3 +135,4 @@
 {"step": 698500, "loss": 0.5305786728858948, "lr": 8e-05, "grad_norm": 0.21394717140636219, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102241679933, "tokens_per_second": 153088.39200416883, "data_wait_s": 1.3371998647926375e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 698600, "loss": 0.5597310066223145, "lr": 8e-05, "grad_norm": 0.20332029918790948, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102256227774, "tokens_per_second": 151777.83470268233, "data_wait_s": 1.676499960012734e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 698700, "loss": 0.5351334810256958, "lr": 8e-05, "grad_norm": 0.2125764499225815, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102270844335, "tokens_per_second": 152332.78818698044, "data_wait_s": 1.2264999895705841e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 698800, "loss": 0.5014955401420593, "lr": 8e-05, "grad_norm": 0.19655496331937625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102285337793, "tokens_per_second": 140519.8405952182, "data_wait_s": 1.6682999557815492e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:02c512319e0dbd7107eb32217d52f53dc8c79282c3dc9c9253918e915d4b595b
-size 291897
+oid sha256:c94dbba493b01d4e85d2cd51d6100bce3abc6fea368b18300ee6e9e8d6826b49
+size 294015