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": 75377.65756947598,
-      "required_pages_per_second_upper_estimate": 0.006533544315195668
+      "required_token_positions_per_second": 74854.88167634227,
+      "required_pages_per_second_upper_estimate": 0.006488231425742225
     },
     {
       "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": 9422.207196184498,
-      "required_pages_per_second_upper_estimate": 0.003758659325109501
+      "required_token_positions_per_second": 9356.860209542783,
+      "required_pages_per_second_upper_estimate": 0.0037325914351136044
     },
     {
       "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": 18844.414392368995,
-      "required_pages_per_second_upper_estimate": 0.004359323619308528
+      "required_token_positions_per_second": 18713.720419085566,
+      "required_pages_per_second_upper_estimate": 0.004329089868724777
     },
     {
       "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": 28266.62158855349,
-      "required_pages_per_second_upper_estimate": 0.001721288951240931
+      "required_token_positions_per_second": 28070.58062862835,
+      "required_pages_per_second_upper_estimate": 0.0017093510853289165
     },
     {
       "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": 28266.62158855349,
-      "required_pages_per_second_upper_estimate": 0.0021760427769034365
+      "required_token_positions_per_second": 28070.58062862835,
+      "required_pages_per_second_upper_estimate": 0.0021609510011322895
     },
     {
       "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": 28266.62158855349,
-      "required_pages_per_second_upper_estimate": 0.35097683783296485
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28070.58062862835,
+      "required_pages_per_second_upper_estimate": 0.34854266521758137
     }
   ],
   "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": 0.12704685807037208,
-    "producer_p95_seconds": 0.3083866710003349,
-    "gpu_update_seconds": 0.9781147673903838,
-    "producer_batches_per_second": 7.871111613370978,
-    "gpu_batches_per_second": 1.0223749127804358,
-    "producer_headroom": 7.698850504816103,
+    "producer_batch_seconds": 0.07897261444520609,
+    "producer_p95_seconds": 0.3008080700019491,
+    "gpu_update_seconds": 0.9849457824111635,
+    "producer_batches_per_second": 12.662617377240743,
+    "gpu_batches_per_second": 1.0152843109312917,
+    "producer_headroom": 12.471991579999578,
     "sustainable": true,
-    "buffer_seconds": 60.6431155782038,
+    "buffer_seconds": 60.08169272708097,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 412973.2507971652,
-    "upload_bytes_per_second": 4114433.733662277,
+    "generation_bytes_per_second": 369447.29963066,
+    "upload_bytes_per_second": 4127713.71209437,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.61444609581201,
+    "estimated_full_shards_per_hour": 79.27479020776605,
     "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": 1.2285513930000889,
-      "bytes": 6249359,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.8377171489992179,
       "bytes": 5823773,
@@ -562,6 +556,12 @@
       "bytes": 11989707,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 0.9998438940019696,
+      "bytes": 6675780,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -207,3 +207,4 @@
 {"step": 105700, "loss": 0.6073123849928379, "lr": 0.0003, "grad_norm": 0.1396080629189812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15467619516, "tokens_per_second": 148694.89898032256, "data_wait_s": 1.371300095343031e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 105800, "loss": 0.5691022537648678, "lr": 0.0003, "grad_norm": 0.13232588957847485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15482144994, "tokens_per_second": 147517.791146581, "data_wait_s": 1.3156997738406062e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 105900, "loss": 0.6305570360273123, "lr": 0.0003, "grad_norm": 0.13164483919808131, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15496873678, "tokens_per_second": 136118.9266714003, "data_wait_s": 1.3787997886538506e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 106000, "loss": 0.6038626618683338, "lr": 0.0003, "grad_norm": 0.13520886709246546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15511593792, "tokens_per_second": 149484.15233918777, "data_wait_s": 1.5450998034793884e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1cf287655437f6ac6d0ff9f19dc4ef247417ff854da151905dc5f16f29144bd7
-size 445959
+oid sha256:8e249ad54ec700a20fcafc32fcd61f71c03b5fd1340443ae786a9ecb55b07dad
+size 448077