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": 76970.16703501086,
-      "required_pages_per_second_upper_estimate": 0.005324974750875216
+      "required_token_positions_per_second": 76824.80162191723,
+      "required_pages_per_second_upper_estimate": 0.00531491803430317
     },
     {
       "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": 9621.270879376358,
-      "required_pages_per_second_upper_estimate": 0.004028225202212115
+      "required_token_positions_per_second": 9603.100202739653,
+      "required_pages_per_second_upper_estimate": 0.004020617519351203
     },
     {
       "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": 19242.541758752715,
-      "required_pages_per_second_upper_estimate": 0.004007713988396808
+      "required_token_positions_per_second": 19206.200405479307,
+      "required_pages_per_second_upper_estimate": 0.004000145042895892
     },
     {
       "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": 28863.81263812907,
-      "required_pages_per_second_upper_estimate": 0.0017303414951658194
+      "required_token_positions_per_second": 28809.300608218957,
+      "required_pages_per_second_upper_estimate": 0.0017270735822077581
     },
     {
       "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": 28863.81263812907,
-      "required_pages_per_second_upper_estimate": 0.0022440344741201275
+      "required_token_positions_per_second": 28809.300608218957,
+      "required_pages_per_second_upper_estimate": 0.002239796403567696
     },
     {
       "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": 28863.81263812907,
-      "required_pages_per_second_upper_estimate": 0.370233996974501
+      "required_token_positions_per_second": 28809.300608218957,
+      "required_pages_per_second_upper_estimate": 0.36953477518527156
     }
   ],
   "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": 0.08791774366375193,
-    "producer_p95_seconds": 0.31667749600455863,
-    "gpu_update_seconds": 0.9578776146667823,
-    "producer_batches_per_second": 11.374268245834148,
-    "gpu_batches_per_second": 1.0439747047934416,
-    "producer_headroom": 10.89515693589974,
+    "producer_batch_seconds": 0.46613480814050945,
+    "producer_p95_seconds": 0.35763232800672995,
+    "gpu_update_seconds": 0.9596900798109742,
+    "producer_batches_per_second": 2.1453021369272314,
+    "gpu_batches_per_second": 1.0420030601931047,
+    "producer_headroom": 2.0588251790063485,
     "sustainable": true,
-    "buffer_seconds": 60.346289724007285,
+    "buffer_seconds": 60.46047502809137,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 262757.1587774979,
-    "upload_bytes_per_second": 3886459.349371184,
+    "generation_bytes_per_second": 268219.6823755835,
+    "upload_bytes_per_second": 3903464.288763667,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.381569600045225,
+    "estimated_full_shards_per_hour": 57.553700003153125,
     "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": 1.2067982320004376,
-      "bytes": 5015589,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.4427611729988712,
       "bytes": 13297372,
@@ -562,6 +556,12 @@
       "bytes": 1202687,
       "pages": 36,
       "shards": 1
+    },
+    {
+      "seconds": 2.338399190994096,
+      "bytes": 11297335,
+      "pages": 36,
+      "shards": 5
     }
   ],
   "cache_key": {

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

@@ -358,3 +358,4 @@
 {"step": 878300, "loss": 0.5400705337524414, "lr": 3e-05, "grad_norm": 0.3049035014871074, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128558742382, "tokens_per_second": 153393.47326460722, "data_wait_s": 1.1251999239902943e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 878400, "loss": 0.5285022258758545, "lr": 3e-05, "grad_norm": 0.3028094354470684, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128573480394, "tokens_per_second": 153739.61314169862, "data_wait_s": 1.648499892326072e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 878500, "loss": 0.5892738699913025, "lr": 3e-05, "grad_norm": 0.3085443776722021, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128588161800, "tokens_per_second": 151940.98682745785, "data_wait_s": 9.24799678614363e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 878600, "loss": 0.5238515138626099, "lr": 3e-05, "grad_norm": 0.3148988540399598, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128602737008, "tokens_per_second": 152306.6153859522, "data_wait_s": 1.9182996766176075e-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 sha256:59618ca47f74fbcd53cc78d7244d4a3f1a480530b482e0ae8dd53e49a4a68f46
-size 768126
+oid sha256:fb453001fb7cc5704543e5cb4a4eaf13b22d7fdce7b6103bcc92475c71cb2ae8
+size 770244