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": 77059.06435411666,
-      "required_pages_per_second_upper_estimate": 0.005331124873681154
+      "required_token_positions_per_second": 77374.67001057314,
+      "required_pages_per_second_upper_estimate": 0.0053529592053008284
     },
     {
       "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": 9632.383044264583,
-      "required_pages_per_second_upper_estimate": 0.0040328776336024256
+      "required_token_positions_per_second": 9671.833751321643,
+      "required_pages_per_second_upper_estimate": 0.004049394820822773
     },
     {
       "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": 19264.766088529166,
-      "required_pages_per_second_upper_estimate": 0.004012342730194254
+      "required_token_positions_per_second": 19343.667502643286,
+      "required_pages_per_second_upper_estimate": 0.004028775813984009
     },
     {
       "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": 28897.14913279375,
-      "required_pages_per_second_upper_estimate": 0.001732339967119087
+      "required_token_positions_per_second": 29015.50125396493,
+      "required_pages_per_second_upper_estimate": 0.0017394349960700736
     },
     {
       "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": 28897.14913279375,
-      "required_pages_per_second_upper_estimate": 0.002246626239948563
+      "required_token_positions_per_second": 29015.50125396493,
+      "required_pages_per_second_upper_estimate": 0.002255827596793657
     },
     {
       "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": 28897.14913279375,
-      "required_pages_per_second_upper_estimate": 0.37066160173412027
+      "required_token_positions_per_second": 29015.50125396493,
+      "required_pages_per_second_upper_estimate": 0.37217969566789716
     }
   ],
   "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": 0.11373688333583232,
-    "producer_p95_seconds": 0.3418942549978965,
-    "gpu_update_seconds": 0.9567725824075787,
-    "producer_batches_per_second": 8.792222634123775,
-    "gpu_batches_per_second": 1.045180451851626,
-    "producer_headroom": 8.412157554752968,
+    "producer_batch_seconds": 0.10960867989098233,
+    "producer_p95_seconds": 0.331680408999091,
+    "gpu_update_seconds": 0.9528699765688845,
+    "producer_batches_per_second": 9.123365056440859,
+    "gpu_batches_per_second": 1.04946112753056,
+    "producer_headroom": 8.693380647560181,
     "sustainable": true,
-    "buffer_seconds": 60.276672691677454,
+    "buffer_seconds": 60.030808523839724,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 525092.6505426638,
-    "upload_bytes_per_second": 4971389.461828683,
+    "generation_bytes_per_second": 529288.4208511035,
+    "upload_bytes_per_second": 4948349.199228604,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 112.67265927514967,
+    "estimated_full_shards_per_hour": 113.57297391080692,
     "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": 1.2786440780000703,
-      "bytes": 7516874,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.9931387870019535,
       "bytes": 6287159,
@@ -562,6 +556,12 @@
       "bytes": 1282728,
       "pages": 37,
       "shards": 1
+    },
+    {
+      "seconds": 1.3358671570022125,
+      "bytes": 5881131,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -368,3 +368,4 @@
 {"step": 477000, "loss": 0.6186416745185852, "lr": 0.0003, "grad_norm": 0.1934919045075562, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69817582261, "tokens_per_second": 151649.48824863945, "data_wait_s": 1.2336000509094447e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 477100, "loss": 0.567695140838623, "lr": 0.0003, "grad_norm": 0.16620838980671304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69832155720, "tokens_per_second": 151627.29816911495, "data_wait_s": 1.6814999980852008e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 477200, "loss": 0.5887116193771362, "lr": 0.0003, "grad_norm": 0.1907135614756541, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69846950325, "tokens_per_second": 153576.297756242, "data_wait_s": 9.424999007023871e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 477300, "loss": 0.6267165541648865, "lr": 0.0003, "grad_norm": 0.15859574941267826, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69861676226, "tokens_per_second": 153357.17745578042, "data_wait_s": 1.3036005839239806e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c34f5e3959f838442d71c3b2b7cbe0ed1b0e2f00f5ce56bd72e5d25ffb45e452
-size 789306
+oid sha256:0b64d4d0bb91235ecf71fcb01bdf805dbef11aa1bb430b3934775fb812de495d
+size 791424