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": 76974.57829516343,
-      "required_pages_per_second_upper_estimate": 0.005325279932088104
+      "required_token_positions_per_second": 76607.31531359084,
+      "required_pages_per_second_upper_estimate": 0.005299871826855389
     },
     {
       "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": 9621.822286895429,
-      "required_pages_per_second_upper_estimate": 0.004028456065025652
+      "required_token_positions_per_second": 9575.914414198855,
+      "required_pages_per_second_upper_estimate": 0.004009235397392992
     },
     {
       "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": 19243.644573790858,
-      "required_pages_per_second_upper_estimate": 0.004007943675686065
+      "required_token_positions_per_second": 19151.82882839771,
+      "required_pages_per_second_upper_estimate": 0.003988820877264707
     },
     {
       "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": 28865.466860686287,
-      "required_pages_per_second_upper_estimate": 0.0017304406632822728
+      "required_token_positions_per_second": 28727.743242596564,
+      "required_pages_per_second_upper_estimate": 0.0017221843426695825
     },
     {
       "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": 28865.466860686287,
-      "required_pages_per_second_upper_estimate": 0.002244163082647795
+      "required_token_positions_per_second": 28727.743242596564,
+      "required_pages_per_second_upper_estimate": 0.002233455677123494
     },
     {
       "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": 28865.466860686287,
-      "required_pages_per_second_upper_estimate": 0.37025521556529917
+      "required_token_positions_per_second": 28727.743242596564,
+      "required_pages_per_second_upper_estimate": 0.36848864486854405
     }
   ],
   "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": 0.10266903780453163,
-    "producer_p95_seconds": 0.2935257479985012,
-    "gpu_update_seconds": 0.9578227206037525,
-    "producer_batches_per_second": 9.740034789299076,
-    "gpu_batches_per_second": 1.0440345363384798,
-    "producer_headroom": 9.329226620661638,
+    "producer_batch_seconds": 0.08002181578137879,
+    "producer_p95_seconds": 0.27273300800152356,
+    "gpu_update_seconds": 0.9624146166484963,
+    "producer_batches_per_second": 12.496592213453642,
+    "gpu_batches_per_second": 1.0390532133462298,
+    "producer_headroom": 12.02690300452357,
     "sustainable": true,
-    "buffer_seconds": 60.34283139803641,
+    "buffer_seconds": 60.632120848855266,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 344489.5146763433,
-    "upload_bytes_per_second": 4809291.121642504,
+    "generation_bytes_per_second": 331985.6448161647,
+    "upload_bytes_per_second": 4771764.764143382,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.91943054406856,
+    "estimated_full_shards_per_hour": 71.2363911472674,
     "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": 1.7856543759990018,
-      "bytes": 17299146,
-      "pages": 27,
-      "shards": 6
-    },
     {
       "seconds": 0.9336543670033279,
       "bytes": 1145946,
@@ -562,6 +556,12 @@
       "bytes": 5944518,
       "pages": 25,
       "shards": 2
+    },
+    {
+      "seconds": 1.3701994980001473,
+      "bytes": 11970289,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -285,3 +285,4 @@
 {"step": 871000, "loss": 0.5876473188400269, "lr": 3e-05, "grad_norm": 0.3304013484529452, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127489898616, "tokens_per_second": 153219.22070710637, "data_wait_s": 1.2127999070798978e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 871100, "loss": 0.5098770260810852, "lr": 3e-05, "grad_norm": 0.31153108474970453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127504582815, "tokens_per_second": 146808.16233404411, "data_wait_s": 1.2897999113192782e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 871200, "loss": 0.553991436958313, "lr": 3e-05, "grad_norm": 0.3167068082192046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127519303992, "tokens_per_second": 152142.7172082724, "data_wait_s": 1.2667002010857686e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 871300, "loss": 0.5202050805091858, "lr": 3e-05, "grad_norm": 0.30355033474078613, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127534016428, "tokens_per_second": 152912.7244680361, "data_wait_s": 1.2471002264646813e-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 sha256:45f7c0e0720595dbb5b8ddb37953204323b022ef8b368e84a536c76361b115be
-size 612729
+oid sha256:7632c1a7bce4e4b87f0be563211174cb5e3917184b5c2f0fe7e740a2483d1733
+size 614847