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": 76148.95662856006,
-      "required_pages_per_second_upper_estimate": 0.005268161509486284
+      "required_token_positions_per_second": 76012.46933056027,
+      "required_pages_per_second_upper_estimate": 0.0052587190015164955
     },
     {
       "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": 9518.619578570007,
-      "required_pages_per_second_upper_estimate": 0.0039852472461674145
+      "required_token_positions_per_second": 9501.558666320034,
+      "required_pages_per_second_upper_estimate": 0.003978104198480712
     },
     {
       "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": 19037.239157140015,
-      "required_pages_per_second_upper_estimate": 0.003964954870674576
+      "required_token_positions_per_second": 19003.117332640068,
+      "required_pages_per_second_upper_estimate": 0.003957848194484262
     },
     {
       "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": 28555.858735710022,
-      "required_pages_per_second_upper_estimate": 0.0017118801289341798
+      "required_token_positions_per_second": 28504.6759989601,
+      "required_pages_per_second_upper_estimate": 0.001708811802017536
     },
     {
       "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": 28555.858735710022,
-      "required_pages_per_second_upper_estimate": 0.0022200924127531102
+      "required_token_positions_per_second": 28504.6759989601,
+      "required_pages_per_second_upper_estimate": 0.0022161131800998704
     },
     {
       "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": 28555.858735710022,
-      "required_pages_per_second_upper_estimate": 0.36628389496940805
+      "required_token_positions_per_second": 28504.6759989601,
+      "required_pages_per_second_upper_estimate": 0.36562737777812115
     }
   ],
   "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": 0.12394676724991882,
-    "producer_p95_seconds": 0.27442925899958936,
-    "gpu_update_seconds": 0.9682076191750201,
-    "producer_batches_per_second": 8.06797968343668,
-    "gpu_batches_per_second": 1.032836325799697,
-    "producer_headroom": 7.811479400852661,
+    "producer_batch_seconds": 0.10023494705487224,
+    "producer_p95_seconds": 0.277848449000885,
+    "gpu_update_seconds": 0.9699461239625613,
+    "producer_batches_per_second": 9.976560365244307,
+    "gpu_batches_per_second": 1.0309850983420175,
+    "producer_headroom": 9.676726056747231,
     "sustainable": true,
-    "buffer_seconds": 60.02887238885125,
+    "buffer_seconds": 60.1366596856788,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 399904.4647289828,
-    "upload_bytes_per_second": 4893649.614304227,
+    "generation_bytes_per_second": 397984.20176623284,
+    "upload_bytes_per_second": 4885942.40336543,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.8101888313495,
+    "estimated_full_shards_per_hour": 85.39814510097732,
     "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": 1.248109399999521,
-      "bytes": 7422482,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.451942487999986,
       "bytes": 6077412,
@@ -562,6 +556,12 @@
       "bytes": 6253390,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.4145583839999745,
+      "bytes": 7569385,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -232,3 +232,4 @@
 {"step": 865700, "loss": 0.5129722952842712, "lr": 3e-05, "grad_norm": 0.30290459662874364, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126714534807, "tokens_per_second": 151665.45865019006, "data_wait_s": 1.8237002223031595e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 865800, "loss": 0.5157815217971802, "lr": 3e-05, "grad_norm": 0.3087268007317333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126729082291, "tokens_per_second": 150663.84339931875, "data_wait_s": 1.9134000467602164e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 865900, "loss": 0.555191159248352, "lr": 3e-05, "grad_norm": 0.32342886147925076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126743758706, "tokens_per_second": 152578.89325919474, "data_wait_s": 9.358998795505613e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 866000, "loss": 0.5319803953170776, "lr": 3e-05, "grad_norm": 0.3136856848495784, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126758284351, "tokens_per_second": 150123.62397090017, "data_wait_s": 1.3885000953450799e-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 sha256:51ef20c693df08a008ed999c910e999d3d2fb5c0b5d3aab2844f2b088126a0be
-size 499692
+oid sha256:8f0c4e6dc1dd66dbd16cd7b08f5c03591cc6d1889029a3a6404622deb8d1c71e
+size 501810