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": 76251.09908781432,
-      "required_pages_per_second_upper_estimate": 0.005275227961820652
+      "required_token_positions_per_second": 76775.59848099282,
+      "required_pages_per_second_upper_estimate": 0.0053115140468470035
     },
     {
       "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": 9531.38738597679,
-      "required_pages_per_second_upper_estimate": 0.003990592860506497
+      "required_token_positions_per_second": 9596.949810124102,
+      "required_pages_per_second_upper_estimate": 0.0040180424783978754
     },
     {
       "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": 19062.77477195358,
-      "required_pages_per_second_upper_estimate": 0.003970273265820794
+      "required_token_positions_per_second": 19193.899620248205,
+      "required_pages_per_second_upper_estimate": 0.003997583113725774
     },
     {
       "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": 28594.16215793037,
-      "required_pages_per_second_upper_estimate": 0.0017141763606103504
+      "required_token_positions_per_second": 28790.849430372306,
+      "required_pages_per_second_upper_estimate": 0.0017259674622691688
     },
     {
       "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": 28594.16215793037,
-      "required_pages_per_second_upper_estimate": 0.0022230703353517934
+      "required_token_positions_per_second": 28790.849430372306,
+      "required_pages_per_second_upper_estimate": 0.002238361905648273
     },
     {
       "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": 28594.16215793037,
-      "required_pages_per_second_upper_estimate": 0.36677521014264014
+      "required_token_positions_per_second": 28790.849430372306,
+      "required_pages_per_second_upper_estimate": 0.369298103287186
     }
   ],
   "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": 0.08805015314032971,
-    "producer_p95_seconds": 0.30202070599625586,
-    "gpu_update_seconds": 0.9669106528561825,
-    "producer_batches_per_second": 11.357163665646924,
-    "gpu_batches_per_second": 1.034221721568662,
-    "producer_headroom": 10.981362534545182,
+    "producer_batch_seconds": 0.10416436226574888,
+    "producer_p95_seconds": 0.2976510629960103,
+    "gpu_update_seconds": 0.9603051159314987,
+    "producer_batches_per_second": 9.60021237828687,
+    "gpu_batches_per_second": 1.0413356998832575,
+    "producer_headroom": 9.21913306089778,
     "sustainable": true,
-    "buffer_seconds": 60.915371129939494,
+    "buffer_seconds": 60.49922230368442,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 405718.89023759763,
-    "upload_bytes_per_second": 5158709.626775249,
+    "generation_bytes_per_second": 424882.9976990195,
+    "upload_bytes_per_second": 5140818.546489874,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.05782919259974,
+    "estimated_full_shards_per_hour": 91.17000053623141,
     "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": 1.2314064579986734,
-      "bytes": 7429916,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.4280758529966988,
       "bytes": 11440670,
@@ -562,6 +556,12 @@
       "bytes": 2355430,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.3224051930010319,
+      "bytes": 6398826,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -352,3 +352,4 @@
 {"step": 755200, "loss": 0.5324469804763794, "lr": 5e-05, "grad_norm": 0.25402884094279243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110540040958, "tokens_per_second": 151674.22945400653, "data_wait_s": 1.3431003026198596e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 755300, "loss": 0.49978014826774597, "lr": 5e-05, "grad_norm": 0.2420624575209652, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110554827045, "tokens_per_second": 153012.19600185988, "data_wait_s": 1.5199002518784255e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 755400, "loss": 0.4990067183971405, "lr": 5e-05, "grad_norm": 0.23418413776953095, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110569395966, "tokens_per_second": 150688.3707369091, "data_wait_s": 1.359100133413449e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 755500, "loss": 0.5137624144554138, "lr": 5e-05, "grad_norm": 0.2398494563840914, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110584181302, "tokens_per_second": 152391.71439916862, "data_wait_s": 1.075599720934406e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cb4589666904e778b18a4dcc8987ff61ae5d1a5244eac0fef18de244af31eef8
-size 755418
+oid sha256:9dde0550825f063aee19cae856cb2c73c771a53f2d001e0dc4a0d9ff8055a1a0
+size 757536