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": 77347.66274047524,
-      "required_pages_per_second_upper_estimate": 0.005351090779689955
+      "required_token_positions_per_second": 77237.26573442125,
+      "required_pages_per_second_upper_estimate": 0.005343453258654794
     },
     {
       "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": 9668.457842559405,
-      "required_pages_per_second_upper_estimate": 0.004047981398321015
+      "required_token_positions_per_second": 9654.658216802656,
+      "required_pages_per_second_upper_estimate": 0.004042203783185619
     },
     {
       "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": 19336.91568511881,
-      "required_pages_per_second_upper_estimate": 0.0040273695884510624
+      "required_token_positions_per_second": 19309.316433605312,
+      "required_pages_per_second_upper_estimate": 0.004021621392202009
     },
     {
       "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": 29005.373527678214,
-      "required_pages_per_second_upper_estimate": 0.0017388278543433277
+      "required_token_positions_per_second": 28963.974650407967,
+      "required_pages_per_second_upper_estimate": 0.0017363460548634017
     },
     {
       "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": 29005.373527678214,
-      "required_pages_per_second_upper_estimate": 0.002255040210622025
+      "required_token_positions_per_second": 28963.974650407967,
+      "required_pages_per_second_upper_estimate": 0.00225182163000868
     },
     {
       "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": 29005.373527678214,
-      "required_pages_per_second_upper_estimate": 0.3720497880693964
+      "required_token_positions_per_second": 28963.974650407967,
+      "required_pages_per_second_upper_estimate": 0.37151876772242487
     }
   ],
   "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": 0.09992235628152457,
-    "producer_p95_seconds": 0.35867528800008586,
-    "gpu_update_seconds": 0.9532026875508792,
-    "producer_batches_per_second": 10.007770405078986,
-    "gpu_batches_per_second": 1.0490948179860464,
-    "producer_headroom": 9.53943364651344,
+    "producer_batch_seconds": 0.11611135932031402,
+    "producer_p95_seconds": 0.35010773099929793,
+    "gpu_update_seconds": 0.9545651221459887,
+    "producer_batches_per_second": 8.612421780726212,
+    "gpu_batches_per_second": 1.0475974627607048,
+    "producer_headroom": 8.22111744909169,
     "sustainable": true,
-    "buffer_seconds": 60.05176931570539,
+    "buffer_seconds": 60.13760269519729,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 272605.1151164689,
-    "upload_bytes_per_second": 4791883.095739071,
+    "generation_bytes_per_second": 282067.8041841313,
+    "upload_bytes_per_second": 4677654.067552161,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.49471178169776,
+    "estimated_full_shards_per_hour": 60.52518457549052,
     "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": 1.4766436999998405,
-      "bytes": 10238885,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.4429230239984463,
       "bytes": 2385665,
@@ -562,6 +556,12 @@
       "bytes": 11121609,
       "pages": 31,
       "shards": 4
+    },
+    {
+      "seconds": 1.7561167999956524,
+      "bytes": 1070087,
+      "pages": 31,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -434,3 +434,4 @@
 {"step": 688400, "loss": 0.5875371098518372, "lr": 0.0001, "grad_norm": 0.213001294053636, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100763397324, "tokens_per_second": 141106.5089613713, "data_wait_s": 1.1366995749995112e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 688500, "loss": 0.5560246706008911, "lr": 0.0001, "grad_norm": 0.18670503836679223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100778019023, "tokens_per_second": 152148.12010560403, "data_wait_s": 1.3844997738488019e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 688600, "loss": 0.5817644596099854, "lr": 0.0001, "grad_norm": 0.20325837312311798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100792624872, "tokens_per_second": 152159.29606921034, "data_wait_s": 1.4031000318937004e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 688700, "loss": 0.5967218279838562, "lr": 0.0001, "grad_norm": 0.20523049831808476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100807133319, "tokens_per_second": 150947.85528205158, "data_wait_s": 1.0051997378468513e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c6fbb8240600378f1571c3940f0eccca647dd23db1c9117ad14ed6ba05cc14f6
-size 929877
+oid sha256:8717049faeaf2d55c597bf1b005eefc54c4a2a0b53f7948333b824e73d3106cd
+size 931995