Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75619.07020211437,
-      "required_pages_per_second_upper_estimate": 0.006554469350338039
+      "required_token_positions_per_second": 74475.84621014274,
+      "required_pages_per_second_upper_estimate": 0.006455377592188657
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9452.383775264296,
-      "required_pages_per_second_upper_estimate": 0.0037706972136844967
+      "required_token_positions_per_second": 9309.480776267843,
+      "required_pages_per_second_upper_estimate": 0.003713691070794576
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18904.76755052859,
-      "required_pages_per_second_upper_estimate": 0.004373285260269445
+      "required_token_positions_per_second": 18618.961552535686,
+      "required_pages_per_second_upper_estimate": 0.004307169072647554
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28357.15132579289,
-      "required_pages_per_second_upper_estimate": 0.0017268017372659868
+      "required_token_positions_per_second": 27928.442328803525,
+      "required_pages_per_second_upper_estimate": 0.0017006956086116096
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28357.15132579289,
-      "required_pages_per_second_upper_estimate": 0.0021830120066784774
+      "required_token_positions_per_second": 27928.442328803525,
+      "required_pages_per_second_upper_estimate": 0.0021500088013477778
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28357.15132579289,
-      "required_pages_per_second_upper_estimate": 0.35210091418593803
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27928.442328803525,
+      "required_pages_per_second_upper_estimate": 0.34677778324004527
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07354231682811019,
-    "producer_p95_seconds": 0.2549473239996587,
-    "gpu_update_seconds": 0.9749921521507746,
-    "producer_batches_per_second": 13.597613498324934,
-    "gpu_batches_per_second": 1.025649281170171,
-    "producer_headroom": 13.25756644884625,
+    "producer_batch_seconds": 0.23584550648433833,
+    "producer_p95_seconds": 0.3246751980004774,
+    "gpu_update_seconds": 0.9899585402758285,
+    "producer_batches_per_second": 4.240063823587864,
+    "gpu_batches_per_second": 1.0101433133971183,
+    "producer_headroom": 4.19748739347539,
     "sustainable": true,
-    "buffer_seconds": 60.449513433348024,
+    "buffer_seconds": 60.38747095682554,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 174,
+    "upload_pages_per_commit": 166,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 615032.5312327559,
-    "upload_bytes_per_second": 4873224.191965345,
+    "generation_bytes_per_second": 579540.4209983335,
+    "upload_bytes_per_second": 4829540.12601437,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.98583198898795,
+    "estimated_full_shards_per_hour": 62.1779416678548,
     "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": 1.3755702170001314,
-      "bytes": 7451679,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.598954668999795,
       "bytes": 1262837,
@@ -562,6 +556,12 @@
       "bytes": 6167341,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.8386335059985868,
+      "bytes": 1124320,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -222,3 +222,4 @@
 {"step": 217200, "loss": 0.5898901894688606, "lr": 0.0003, "grad_norm": 0.15459375768144146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31791071925, "tokens_per_second": 148009.81449124365, "data_wait_s": 1.0289000783814117e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 217300, "loss": 0.5165498238056898, "lr": 0.0003, "grad_norm": 0.19917937040410089, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31805678264, "tokens_per_second": 148768.6150525741, "data_wait_s": 1.1164000170538202e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 217400, "loss": 0.5930443629622459, "lr": 0.0003, "grad_norm": 0.14338717620554772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31820352549, "tokens_per_second": 148361.99374354762, "data_wait_s": 8.110000635497272e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 217500, "loss": 0.6024927664548159, "lr": 0.0003, "grad_norm": 0.158279336983577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31835045908, "tokens_per_second": 148549.87263134716, "data_wait_s": 1.2456999684218317e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6ec6e3aaeaa7cf0b72456b2f06d51fca2669b171ae9f9c272b9cdb2fbfe23aed
-size 477729
+oid sha256:784fd5835549bb7e5781a7fad82400f81d0733425da6638dc7d5059e6fbc748c
+size 479847