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": 75736.8819775496,
-      "required_pages_per_second_upper_estimate": 0.00656468097644155
+      "required_token_positions_per_second": 72521.81026857447,
+      "required_pages_per_second_upper_estimate": 0.0062860067092323346
     },
     {
       "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": 9467.1102471937,
-      "required_pages_per_second_upper_estimate": 0.0037765718235175123
+      "required_token_positions_per_second": 9065.226283571808,
+      "required_pages_per_second_upper_estimate": 0.003616254301728023
     },
     {
       "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": 18934.2204943874,
-      "required_pages_per_second_upper_estimate": 0.004380098680477072
+      "required_token_positions_per_second": 18130.452567143617,
+      "required_pages_per_second_upper_estimate": 0.004194161116341602
     },
     {
       "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": 28401.330741581096,
-      "required_pages_per_second_upper_estimate": 0.0017294920318960078
+      "required_token_positions_per_second": 27195.678850715427,
+      "required_pages_per_second_upper_estimate": 0.0016560741573088944
     },
     {
       "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": 28401.330741581096,
-      "required_pages_per_second_upper_estimate": 0.002186413060402304
+      "required_token_positions_per_second": 27195.678850715427,
+      "required_pages_per_second_upper_estimate": 0.002093598640385424
     },
     {
       "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": 28401.330741581096,
-      "required_pages_per_second_upper_estimate": 0.35264947467103436
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27195.678850715427,
+      "required_pages_per_second_upper_estimate": 0.33767931324379385
     }
   ],
   "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": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11757013041409436,
-    "producer_p95_seconds": 0.39260148700122954,
-    "gpu_update_seconds": 0.9734755125231446,
-    "producer_batches_per_second": 8.505561714339304,
-    "gpu_batches_per_second": 1.0272472056416775,
-    "producer_headroom": 8.27995604916369,
+    "producer_batch_seconds": 0.42578782423424855,
+    "producer_p95_seconds": 0.3704886740015354,
+    "gpu_update_seconds": 1.0166320962888071,
+    "producer_batches_per_second": 2.3485875900712623,
+    "gpu_batches_per_second": 0.9836400047278439,
+    "producer_headroom": 2.387649525012025,
     "sustainable": true,
-    "buffer_seconds": 60.355481776434964,
+    "buffer_seconds": 60.997925777328426,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 298588.58826681913,
-    "upload_bytes_per_second": 5503103.00585859,
+    "generation_bytes_per_second": 275587.27278525,
+    "upload_bytes_per_second": 5494949.8538199235,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.07016025546484,
+    "estimated_full_shards_per_hour": 59.134613396340605,
     "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": 1.0202759780004271,
-      "bytes": 6615812,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.4830753700007335,
       "bytes": 9974834,
@@ -562,6 +556,12 @@
       "bytes": 2271023,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.0174418120004702,
+      "bytes": 5974488,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -352,3 +352,4 @@
 {"step": 150200, "loss": 0.6136030592024326, "lr": 0.0003, "grad_norm": 0.15841969749570134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21981372798, "tokens_per_second": 141381.67709633234, "data_wait_s": 1.2587996025104076e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 150300, "loss": 0.6305031254887581, "lr": 0.0003, "grad_norm": 0.13981532425544294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21995994782, "tokens_per_second": 149911.07402587374, "data_wait_s": 1.2126001820433885e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 150400, "loss": 0.5901217982172966, "lr": 0.0003, "grad_norm": 0.13779970108432205, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22010611914, "tokens_per_second": 145315.96758618526, "data_wait_s": 1.0153999028261751e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 150500, "loss": 0.6060477271676064, "lr": 0.0003, "grad_norm": 0.13552522216887578, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22025162642, "tokens_per_second": 148150.53943653347, "data_wait_s": 8.285998774226755e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0b07fadf572750a068a1531df0485eefb5493dbc0fdee52a4e4b76cf56b1ee25
-size 755418
+oid sha256:b65daa5657c50bc810187fb473a0cfbd493a9a66d4fe099055536b6e4b985c17
+size 757536