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": 76180.06370583098,
-      "required_pages_per_second_upper_estimate": 0.005270313569270249
+      "required_token_positions_per_second": 76459.8491244812,
+      "required_pages_per_second_upper_estimate": 0.005289669773724084
     },
     {
       "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": 9522.507963228873,
-      "required_pages_per_second_upper_estimate": 0.003986875231625376
+      "required_token_positions_per_second": 9557.48114056015,
+      "required_pages_per_second_upper_estimate": 0.004001517770650991
     },
     {
       "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": 19045.015926457745,
-      "required_pages_per_second_upper_estimate": 0.003966574566636261
+      "required_token_positions_per_second": 19114.9622811203,
+      "required_pages_per_second_upper_estimate": 0.003981142547703055
     },
     {
       "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": 28567.523889686618,
-      "required_pages_per_second_upper_estimate": 0.001712579437103418
+      "required_token_positions_per_second": 28672.44342168045,
+      "required_pages_per_second_upper_estimate": 0.001718869202843599
     },
     {
       "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": 28567.523889686618,
-      "required_pages_per_second_upper_estimate": 0.0022209993271652007
+      "required_token_positions_per_second": 28672.44342168045,
+      "required_pages_per_second_upper_estimate": 0.002229156359285472
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28567.523889686618,
-      "required_pages_per_second_upper_estimate": 0.36643352303955334
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28672.44342168045,
+      "required_pages_per_second_upper_estimate": 0.3677793181421538
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1412275848903164,
-    "producer_p95_seconds": 0.3398327530012466,
-    "gpu_update_seconds": 0.9678122649608221,
-    "producer_batches_per_second": 7.080769672416649,
-    "gpu_batches_per_second": 1.0332582425378551,
-    "producer_headroom": 6.852855734327455,
+    "producer_batch_seconds": 0.3208995123203522,
+    "producer_p95_seconds": 0.3169216980022611,
+    "gpu_update_seconds": 0.9642708015283474,
+    "producer_batches_per_second": 3.116240323237717,
+    "gpu_batches_per_second": 1.0370530751475857,
+    "producer_headroom": 3.00489955424339,
     "sustainable": true,
-    "buffer_seconds": 60.004360427570965,
+    "buffer_seconds": 60.749060496285885,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 263436.9195044395,
-    "upload_bytes_per_second": 4687047.1591097815,
+    "generation_bytes_per_second": 258255.29126927638,
+    "upload_bytes_per_second": 4707891.815556258,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.52743042802705,
+    "estimated_full_shards_per_hour": 55.41557363089293,
     "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": 1.1668012100017222,
-      "bytes": 6253390,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.4145583839999745,
       "bytes": 7569385,
@@ -562,6 +556,12 @@
       "bytes": 5765469,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.2353659159998642,
+      "bytes": 8486468,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -310,3 +310,4 @@
 {"step": 873500, "loss": 0.5143589973449707, "lr": 3e-05, "grad_norm": 0.29719932330998033, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127855896756, "tokens_per_second": 151729.69983282857, "data_wait_s": 1.980800152523443e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 873600, "loss": 0.5204393267631531, "lr": 3e-05, "grad_norm": 0.30090023893855966, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127870633405, "tokens_per_second": 152781.50109899126, "data_wait_s": 1.0993000614689663e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 873700, "loss": 0.5502350926399231, "lr": 3e-05, "grad_norm": 0.3079950430755203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127885308473, "tokens_per_second": 152757.36369143435, "data_wait_s": 1.617399902897887e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 873800, "loss": 0.5340554118156433, "lr": 3e-05, "grad_norm": 0.30434055765120077, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127899912309, "tokens_per_second": 151690.4765906996, "data_wait_s": 9.535000572213903e-06, "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 sha256:3faa268f42419c504c7fcc13fc7981c445cfb4d021173b4f7b94cfa704499166
-size 665679
+oid sha256:a44ed7ef3e0a4086f28ab5060fb3712edd35718783dd32c13c3e8b3c071a5c05
+size 667797