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": 76575.42380352163,
-      "required_pages_per_second_upper_estimate": 0.005297665498190303
+      "required_token_positions_per_second": 76325.21520516719,
+      "required_pages_per_second_upper_estimate": 0.005280355486792209
     },
     {
       "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": 9571.927975440203,
-      "required_pages_per_second_upper_estimate": 0.0040075663587310515
+      "required_token_positions_per_second": 9540.651900645898,
+      "required_pages_per_second_upper_estimate": 0.0039944717193333866
     },
     {
       "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": 19143.855950880406,
-      "required_pages_per_second_upper_estimate": 0.0039871603371367725
+      "required_token_positions_per_second": 19081.303801291797,
+      "required_pages_per_second_upper_estimate": 0.003974132373988581
     },
     {
       "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": 28715.78392632061,
-      "required_pages_per_second_upper_estimate": 0.0017214673999196577
+      "required_token_positions_per_second": 28621.9557019377,
+      "required_pages_per_second_upper_estimate": 0.0017158425411353054
     },
     {
       "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": 28715.78392632061,
-      "required_pages_per_second_upper_estimate": 0.0022325258928865117
+      "required_token_positions_per_second": 28621.9557019377,
+      "required_pages_per_second_upper_estimate": 0.002225231161147483
     },
     {
       "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": 28715.78392632061,
-      "required_pages_per_second_upper_estimate": 0.36833524360026954
+      "required_token_positions_per_second": 28621.9557019377,
+      "required_pages_per_second_upper_estimate": 0.3671317158827837
     }
   ],
   "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": 0.0851737083124533,
-    "producer_p95_seconds": 0.23203550600010203,
-    "gpu_update_seconds": 0.9628154352651369,
-    "producer_batches_per_second": 11.740712243402339,
-    "gpu_batches_per_second": 1.0386206570573138,
-    "producer_headroom": 11.304138968954145,
+    "producer_batch_seconds": 0.06221374746100139,
+    "producer_p95_seconds": 0.20255586900020717,
+    "gpu_update_seconds": 0.9659717277155956,
+    "producer_batches_per_second": 16.073617822601808,
+    "gpu_batches_per_second": 1.0352269857471679,
+    "producer_headroom": 15.526660378738859,
     "sustainable": true,
-    "buffer_seconds": 60.65737242170363,
+    "buffer_seconds": 60.85621884608252,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 358684.0023050451,
-    "upload_bytes_per_second": 5096264.351656916,
+    "generation_bytes_per_second": 344181.326523725,
+    "upload_bytes_per_second": 5166307.022636873,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.96523715842737,
+    "estimated_full_shards_per_hour": 73.8533005407697,
     "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": 2.7728241000004346,
-      "bytes": 13625242,
-      "pages": 41,
-      "shards": 4
-    },
     {
       "seconds": 3.518534920000093,
       "bytes": 6664316,
@@ -562,6 +556,12 @@
       "bytes": 2228503,
       "pages": 22,
       "shards": 2
+    },
+    {
+      "seconds": 1.1276412609986437,
+      "bytes": 11020284,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -115,3 +115,4 @@
 {"step": 408600, "loss": 0.5965887904167175, "lr": 0.0003, "grad_norm": 0.16157135267235148, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59808262489, "tokens_per_second": 152413.77060938033, "data_wait_s": 1.0130999726243317e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 408700, "loss": 0.5964803695678711, "lr": 0.0003, "grad_norm": 0.1598428420775088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59822867430, "tokens_per_second": 144949.2339805092, "data_wait_s": 7.458000254700892e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 408800, "loss": 0.5853198170661926, "lr": 0.0003, "grad_norm": 0.1617492643326646, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59837481465, "tokens_per_second": 151964.27694449786, "data_wait_s": 1.58220009325305e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 408900, "loss": 0.5690162181854248, "lr": 0.0003, "grad_norm": 0.14205086280354315, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59852127487, "tokens_per_second": 152803.61247731256, "data_wait_s": 1.3678998584509827e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fdbe557c8bfeee201a6c749cb6fd8705e3462338563e133f787d30f7359e866e
-size 249537
+oid sha256:6d8935d3d67655723d4abe426da5981911db109ca506b2a8a4fd1d4cb6954116
+size 251655