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": 76738.88823189473,
-      "required_pages_per_second_upper_estimate": 0.005308974346634884
+      "required_token_positions_per_second": 76388.670678143,
+      "required_pages_per_second_upper_estimate": 0.005284745483649658
     },
     {
       "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": 9592.361028986841,
-      "required_pages_per_second_upper_estimate": 0.004016121251560351
+      "required_token_positions_per_second": 9548.583834767875,
+      "required_pages_per_second_upper_estimate": 0.0039977926545126385
     },
     {
       "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": 19184.722057973682,
-      "required_pages_per_second_upper_estimate": 0.003995671669532585
+      "required_token_positions_per_second": 19097.16766953575,
+      "required_pages_per_second_upper_estimate": 0.003977436399385457
     },
     {
       "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": 28777.083086960527,
-      "required_pages_per_second_upper_estimate": 0.001725142191001622
+      "required_token_positions_per_second": 28645.751504303626,
+      "required_pages_per_second_upper_estimate": 0.0017172690631530561
     },
     {
       "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": 28777.083086960527,
-      "required_pages_per_second_upper_estimate": 0.002237291632999752
+      "required_token_positions_per_second": 28645.751504303626,
+      "required_pages_per_second_upper_estimate": 0.002227081180114761
     },
     {
       "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": 28777.083086960527,
-      "required_pages_per_second_upper_estimate": 0.3691215234150476
+      "required_token_positions_per_second": 28645.751504303626,
+      "required_pages_per_second_upper_estimate": 0.36743694288559187
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06235414972655917,
-    "producer_p95_seconds": 0.22041695000007167,
-    "gpu_update_seconds": 0.9607645054382827,
-    "producer_batches_per_second": 16.03742500515662,
-    "gpu_batches_per_second": 1.0408377852633293,
-    "producer_headroom": 15.408188703582852,
+    "producer_batch_seconds": 0.11039796831252602,
+    "producer_p95_seconds": 0.22677269100313424,
+    "gpu_update_seconds": 0.965169302535012,
+    "producer_batches_per_second": 9.058137711095338,
+    "gpu_batches_per_second": 1.036087655682278,
+    "producer_headroom": 8.742636456883977,
     "sustainable": true,
-    "buffer_seconds": 60.52816384261181,
+    "buffer_seconds": 60.80566605970576,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 345108.13944355503,
+    "generation_bytes_per_second": 346076.79269388673,
     "upload_bytes_per_second": 1896617.331969005,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.05217301826465,
+    "estimated_full_shards_per_hour": 74.26002345669224,
     "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": [

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -185,3 +185,4 @@
 {"step": 663500, "loss": 0.515264630317688, "lr": 0.0001, "grad_norm": 0.1931068198266987, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97119043307, "tokens_per_second": 153308.1898406546, "data_wait_s": 1.1003001418430358e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 663600, "loss": 0.5422067642211914, "lr": 0.0001, "grad_norm": 0.251312046759925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97133724836, "tokens_per_second": 152555.63012917296, "data_wait_s": 1.308799983235076e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 663700, "loss": 0.5779117345809937, "lr": 0.0001, "grad_norm": 0.2095624644082802, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97148526991, "tokens_per_second": 154575.90331443175, "data_wait_s": 1.5607998648192734e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 663800, "loss": 0.5807338356971741, "lr": 0.0001, "grad_norm": 0.1891549641286253, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97163369344, "tokens_per_second": 154110.84281717698, "data_wait_s": 1.8357000953983516e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a8864e82e311645a30fad52e8fac842a118573d26f15866c0c39db0f79d5bd69
-size 398580
+oid sha256:05c51de6c38c8ba68e395078f957012669fada73d2264efc911434e8785248ec
+size 400698