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": 77379.51258991253,
-      "required_pages_per_second_upper_estimate": 0.005353294226175859
+      "required_token_positions_per_second": 77184.6143583517,
+      "required_pages_per_second_upper_estimate": 0.005339810714290279
     },
     {
       "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": 9672.439073739066,
-      "required_pages_per_second_upper_estimate": 0.0040496482566783785
+      "required_token_positions_per_second": 9648.076794793962,
+      "required_pages_per_second_upper_estimate": 0.0040394482792262985
     },
     {
       "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": 19344.87814747813,
-      "required_pages_per_second_upper_estimate": 0.0040290279593762565
+      "required_token_positions_per_second": 19296.153589587924,
+      "required_pages_per_second_upper_estimate": 0.004018879918920727
     },
     {
       "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": 29017.3172212172,
-      "required_pages_per_second_upper_estimate": 0.0017395438605340258
+      "required_token_positions_per_second": 28944.230384381885,
+      "required_pages_per_second_upper_estimate": 0.0017351624162628867
     },
     {
       "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": 29017.3172212172,
-      "required_pages_per_second_upper_estimate": 0.0022559687802599254
+      "required_token_positions_per_second": 28944.230384381885,
+      "required_pages_per_second_upper_estimate": 0.0022502866001710004
     },
     {
       "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": 29017.3172212172,
-      "required_pages_per_second_upper_estimate": 0.37220298894597553
+      "required_token_positions_per_second": 28944.230384381885,
+      "required_pages_per_second_upper_estimate": 0.37126550947758347
     }
   ],
   "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": 0.06791277025794784,
-    "producer_p95_seconds": 0.2400120929996774,
-    "gpu_update_seconds": 0.9528103438792072,
-    "producer_batches_per_second": 14.724771146895893,
-    "gpu_batches_per_second": 1.0495268092164787,
-    "producer_headroom": 14.029914260016504,
+    "producer_batch_seconds": 0.4285058832109314,
+    "producer_p95_seconds": 0.2461003369971877,
+    "gpu_update_seconds": 0.9552162774007865,
+    "producer_batches_per_second": 2.3336902459930786,
+    "gpu_batches_per_second": 1.046883332768442,
+    "producer_headroom": 2.2291789093840344,
     "sustainable": true,
-    "buffer_seconds": 60.02705166439005,
+    "buffer_seconds": 60.17862547624955,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 406840.7278810349,
-    "upload_bytes_per_second": 5326018.8035889305,
+    "generation_bytes_per_second": 371926.69448712945,
+    "upload_bytes_per_second": 5393589.597304088,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.2985494358376,
+    "estimated_full_shards_per_hour": 79.80681062660611,
     "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": 1.1853528239989828,
-      "bytes": 5804023,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.138313076000486,
       "bytes": 10599364,
@@ -562,6 +556,12 @@
       "bytes": 778515,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.26428677900185,
+      "bytes": 11594708,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -229,3 +229,4 @@
 {"step": 580000, "loss": 0.6204338073730469, "lr": 0.0003, "grad_norm": 0.17000273589656398, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84894863938, "tokens_per_second": 152097.15350274555, "data_wait_s": 1.1026997526641935e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 580100, "loss": 0.5881728529930115, "lr": 0.0003, "grad_norm": 0.16651981878891065, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84909566576, "tokens_per_second": 140841.17921632898, "data_wait_s": 8.233000698965043e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 580200, "loss": 0.596977949142456, "lr": 0.0003, "grad_norm": 0.22100880422977204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84924218816, "tokens_per_second": 153454.72095366375, "data_wait_s": 1.2499000149546191e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 580300, "loss": 0.5834414958953857, "lr": 0.0003, "grad_norm": 0.159024908062277, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84938873882, "tokens_per_second": 142447.16291871038, "data_wait_s": 1.0414998541818932e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4e3da52ed822a878b6dc09b563bffbfe024fe1cffd24be44d811829404be5128
-size 493338
+oid sha256:ac0b61010105711802aef72f3566e7a317217a282fabfac34388e50ed17a1687
+size 495456