Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75222.06768658264,
-      "required_pages_per_second_upper_estimate": 0.006520058178485426
+      "required_token_positions_per_second": 75108.18443669063,
+      "required_pages_per_second_upper_estimate": 0.006510187067019244
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9402.75846082283,
-      "required_pages_per_second_upper_estimate": 0.0037509009337892254
+      "required_token_positions_per_second": 9388.523054586329,
+      "required_pages_per_second_upper_estimate": 0.0037452222174031947
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18805.51692164566,
-      "required_pages_per_second_upper_estimate": 0.0043503253740287875
+      "required_token_positions_per_second": 18777.046109172657,
+      "required_pages_per_second_upper_estimate": 0.004343739152632344
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28208.275382468488,
-      "required_pages_per_second_upper_estimate": 0.0017177359734092395
+      "required_token_positions_per_second": 28165.569163758984,
+      "required_pages_per_second_upper_estimate": 0.0017151353887520438
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28208.275382468488,
-      "required_pages_per_second_upper_estimate": 0.0021715511244463775
+      "required_token_positions_per_second": 28165.569163758984,
+      "required_pages_per_second_upper_estimate": 0.0021682634815117403
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28208.275382468488,
-      "required_pages_per_second_upper_estimate": 0.3502523732255794
+      "required_token_positions_per_second": 28165.569163758984,
+      "required_pages_per_second_upper_estimate": 0.34972210491772704
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08970300395310637,
-    "producer_p95_seconds": 0.1373318339999514,
-    "gpu_update_seconds": 0.9801379072321202,
-    "producer_batches_per_second": 11.147898687124965,
-    "gpu_batches_per_second": 1.0202645899330327,
-    "producer_headroom": 10.926478089234363,
+    "producer_batch_seconds": 0.05937797686717339,
+    "producer_p95_seconds": 0.13977028799990876,
+    "gpu_update_seconds": 0.9816240474051933,
+    "producer_batches_per_second": 16.84126089773937,
+    "gpu_batches_per_second": 1.0187199494993846,
+    "producer_headroom": 16.531786685845738,
     "sustainable": true,
-    "buffer_seconds": 60.768550248391456,
+    "buffer_seconds": 60.86069093912198,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 60,
+    "upload_pages_per_commit": 66,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 441036.71321028966,
-    "upload_bytes_per_second": 4337593.560607531,
+    "generation_bytes_per_second": 439064.55884130055,
+    "upload_bytes_per_second": 4589757.733404302,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.63621184569853,
+    "estimated_full_shards_per_hour": 94.21303342751753,
     "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": [
@@ -232,6 +232,12 @@
       "bytes": 6527222,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 0.8425330919999396,
+      "bytes": 6207711,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -7,3 +7,4 @@
 {"step": 82600, "loss": 0.5864890031516552, "lr": 0.0003, "grad_norm": 0.12823567639076558, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12088108413, "tokens_per_second": 147056.68414229367, "data_wait_s": 7.895999942775234e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 82700, "loss": 0.6107049584388733, "lr": 0.0003, "grad_norm": 0.1509346911072232, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12102595000, "tokens_per_second": 147294.77227946167, "data_wait_s": 9.600999874237459e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 82800, "loss": 0.5665384028106928, "lr": 0.0003, "grad_norm": 0.13098018386505245, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12117034475, "tokens_per_second": 145526.04640431632, "data_wait_s": 1.0684000017135986e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 82900, "loss": 0.5804585367441177, "lr": 0.0003, "grad_norm": 0.13427587039895614, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12131681070, "tokens_per_second": 149314.93150573867, "data_wait_s": 1.162500007012568e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a94b9246c3c5c81baf79fbdd31a3717d2ce892fbd52f7937d6ccca95f3b337e2
-size 19227
+oid sha256:780864437174607718dc3840525575aec56ba413bfbe8b748804618fecbe3c35
+size 21345