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": 76591.65130324331,
-      "required_pages_per_second_upper_estimate": 0.005298788154274037
+      "required_token_positions_per_second": 77200.04482094299,
+      "required_pages_per_second_upper_estimate": 0.00534087822949595
     },
     {
       "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": 9573.956412905414,
-      "required_pages_per_second_upper_estimate": 0.004008415623139145
+      "required_token_positions_per_second": 9650.005602617874,
+      "required_pages_per_second_upper_estimate": 0.0040402558307840826
     },
     {
       "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": 19147.91282581083,
-      "required_pages_per_second_upper_estimate": 0.003988005277197793
+      "required_token_positions_per_second": 19300.011205235747,
+      "required_pages_per_second_upper_estimate": 0.004019683358528003
     },
     {
       "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": 28721.869238716245,
-      "required_pages_per_second_upper_estimate": 0.001721832205106042
+      "required_token_positions_per_second": 28950.01680785362,
+      "required_pages_per_second_upper_estimate": 0.0017355093035146601
     },
     {
       "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": 28721.869238716245,
-      "required_pages_per_second_upper_estimate": 0.0022329989991588116
+      "required_token_positions_per_second": 28950.01680785362,
+      "required_pages_per_second_upper_estimate": 0.0022507364691441408
     },
     {
       "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": 28721.869238716245,
-      "required_pages_per_second_upper_estimate": 0.3684132994537813
+      "required_token_positions_per_second": 28950.01680785362,
+      "required_pages_per_second_upper_estimate": 0.3713397315049014
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.09392247097662221,
-    "producer_p95_seconds": 0.272669539997878,
-    "gpu_update_seconds": 0.9626114432249869,
-    "producer_batches_per_second": 10.647079336838415,
-    "gpu_batches_per_second": 1.0388407566086604,
-    "producer_headroom": 10.249000406564964,
+    "producer_batch_seconds": 0.1256485392344473,
+    "producer_p95_seconds": 0.2709942260007665,
+    "gpu_update_seconds": 0.9550253522650666,
+    "producer_batches_per_second": 7.9587077262721095,
+    "gpu_batches_per_second": 1.047092621811835,
+    "producer_headroom": 7.600767649857729,
     "sustainable": true,
-    "buffer_seconds": 60.64452092317418,
+    "buffer_seconds": 60.16659719269919,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 437972.6620339979,
-    "upload_bytes_per_second": 5877535.721379493,
+    "generation_bytes_per_second": 427786.10838930955,
+    "upload_bytes_per_second": 5884017.620535921,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.97873779072717,
+    "estimated_full_shards_per_hour": 91.79294050940958,
     "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": 1.003430278000451,
-      "bytes": 5945767,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.8994475729996338,
       "bytes": 6104947,
@@ -562,6 +556,12 @@
       "bytes": 1077995,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 0.9089211409991549,
+      "bytes": 5857565,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -275,3 +275,4 @@
 {"step": 424600, "loss": 0.5884664058685303, "lr": 0.0003, "grad_norm": 0.1666712249518507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62149745642, "tokens_per_second": 150856.40673508198, "data_wait_s": 1.9003000488737598e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 424700, "loss": 0.589414656162262, "lr": 0.0003, "grad_norm": 0.15922639864515833, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62164424757, "tokens_per_second": 152461.7311451928, "data_wait_s": 8.151997462846339e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 424800, "loss": 0.566508412361145, "lr": 0.0003, "grad_norm": 0.17545006554065132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62179068566, "tokens_per_second": 151526.20415752192, "data_wait_s": 1.5088000509422272e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 424900, "loss": 0.5948567390441895, "lr": 0.0003, "grad_norm": 0.1873019857651058, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62193678720, "tokens_per_second": 138619.7344759474, "data_wait_s": 1.3924000086262822e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a56bfd92fd4b9c3f98c6fc2bac7d7bd605f5aa8a354eac114aac64f6b1f43126
-size 590766
+oid sha256:41589f9c9cfd8aaa3b396e3c8e61bce0c8fe255ddfe704e51bf8254a39bcffbf
+size 592884