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": 75919.41622421025,
-      "required_pages_per_second_upper_estimate": 0.005252281371706247
+      "required_token_positions_per_second": 76099.58692945373,
+      "required_pages_per_second_upper_estimate": 0.005264745999148624
     },
     {
       "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": 9489.92702802628,
-      "required_pages_per_second_upper_estimate": 0.003973234274423345
+      "required_token_positions_per_second": 9512.448366181716,
+      "required_pages_per_second_upper_estimate": 0.003982663488410006
     },
     {
       "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": 18979.85405605256,
-      "required_pages_per_second_upper_estimate": 0.003953003067464943
+      "required_token_positions_per_second": 19024.89673236343,
+      "required_pages_per_second_upper_estimate": 0.003962384269085244
     },
     {
       "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": 28469.781084078844,
-      "required_pages_per_second_upper_estimate": 0.0017067199051518545
+      "required_token_positions_per_second": 28537.34509854515,
+      "required_pages_per_second_upper_estimate": 0.001710770264654833
     },
     {
       "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": 28469.781084078844,
-      "required_pages_per_second_upper_estimate": 0.0022134002539543622
+      "required_token_positions_per_second": 28537.34509854515,
+      "required_pages_per_second_upper_estimate": 0.002218653058896425
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28469.781084078844,
-      "required_pages_per_second_upper_estimate": 0.36517978327726486
+      "required_token_positions_per_second": 28537.34509854515,
+      "required_pages_per_second_upper_estimate": 0.3660464219102519
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08440952440635385,
-    "producer_p95_seconds": 0.22975776299972495,
-    "gpu_update_seconds": 0.9711349700353541,
-    "producer_batches_per_second": 11.84700431654992,
-    "gpu_batches_per_second": 1.0297229848118796,
-    "producer_headroom": 11.505040181961418,
+    "producer_batch_seconds": 0.1409992348671807,
+    "producer_p95_seconds": 0.23499341000024287,
+    "gpu_update_seconds": 0.9688357450395592,
+    "producer_batches_per_second": 7.0922370673995925,
+    "gpu_batches_per_second": 1.0321667063999258,
+    "producer_headroom": 6.871212783191263,
     "sustainable": true,
-    "buffer_seconds": 60.21036814219195,
+    "buffer_seconds": 60.06781619245267,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 392063.53196288,
-    "upload_bytes_per_second": 4699528.373898366,
+    "generation_bytes_per_second": 404896.68523751537,
+    "upload_bytes_per_second": 4829191.667563001,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.12770718731689,
+    "estimated_full_shards_per_hour": 86.88140313953491,
     "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": 1.384273941000174,
-      "bytes": 1166498,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.245850141000119,
       "bytes": 7339507,
@@ -562,6 +556,12 @@
       "bytes": 5473896,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.548700262001148,
+      "bytes": 13538856,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -137,3 +137,4 @@
 {"step": 856200, "loss": 0.49121734499931335, "lr": 3e-05, "grad_norm": 0.3282350773532133, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125323702384, "tokens_per_second": 150488.83834897677, "data_wait_s": 2.2682999770040624e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 856300, "loss": 0.4934995770454407, "lr": 3e-05, "grad_norm": 0.3060387590311734, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125338262747, "tokens_per_second": 150914.3308055724, "data_wait_s": 1.5229001292027533e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 856400, "loss": 0.5333918333053589, "lr": 3e-05, "grad_norm": 0.2950053123149288, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125352737696, "tokens_per_second": 150735.05077462707, "data_wait_s": 1.5937999705784023e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 856500, "loss": 0.5582417845726013, "lr": 3e-05, "grad_norm": 0.2987068590152231, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125367425236, "tokens_per_second": 152650.14897634267, "data_wait_s": 1.9638999219750986e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d65b5492195e6aa5962d7e0913dd1aae7cd1ab571e62e8f29eb235725c5d1fcb
-size 296916
+oid sha256:f4a6d4ae2b58b5f5aebbdd307dcebb42deff7aa45e86c921d9166f51dedff655
+size 299034