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": 76185.47042116868,
-      "required_pages_per_second_upper_estimate": 0.00527068761838262
+      "required_token_positions_per_second": 76168.7507883194,
+      "required_pages_per_second_upper_estimate": 0.005269530915387214
     },
     {
       "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": 9523.183802646085,
-      "required_pages_per_second_upper_estimate": 0.003987158191476231
+      "required_token_positions_per_second": 9521.093848539926,
+      "required_pages_per_second_upper_estimate": 0.003986283171335187
     },
     {
       "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": 19046.36760529217,
-      "required_pages_per_second_upper_estimate": 0.003966856085691303
+      "required_token_positions_per_second": 19042.18769707985,
+      "required_pages_per_second_upper_estimate": 0.00396598552104227
     },
     {
       "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": 28569.551407938252,
-      "required_pages_per_second_upper_estimate": 0.0017127009837267623
+      "required_token_positions_per_second": 28563.281545619775,
+      "required_pages_per_second_upper_estimate": 0.001712325115054295
     },
     {
       "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": 28569.551407938252,
-      "required_pages_per_second_upper_estimate": 0.002221156957791155
+      "required_token_positions_per_second": 28563.281545619775,
+      "required_pages_per_second_upper_estimate": 0.002220669503574103
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28569.551407938252,
-      "required_pages_per_second_upper_estimate": 0.3664595298667058
+      "required_token_positions_per_second": 28563.281545619775,
+      "required_pages_per_second_upper_estimate": 0.36637910680493807
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.5897162843910451,
-    "producer_p95_seconds": 0.32972574900486507,
-    "gpu_update_seconds": 0.9677435814521682,
-    "producer_batches_per_second": 1.6957306868889055,
-    "gpu_batches_per_second": 1.0333315758079518,
-    "producer_headroom": 1.6410324881082148,
+    "producer_batch_seconds": 0.08894884732029595,
+    "producer_p95_seconds": 0.31278362500597723,
+    "gpu_update_seconds": 0.9679560086904603,
+    "producer_batches_per_second": 11.24241662625598,
+    "gpu_batches_per_second": 1.0331048012738635,
+    "producer_headroom": 10.88216472558601,
     "sustainable": true,
-    "buffer_seconds": 60.00010205003443,
+    "buffer_seconds": 60.013272538808536,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 239980.14378523207,
-    "upload_bytes_per_second": 4716010.220967315,
+    "generation_bytes_per_second": 230507.31512107616,
+    "upload_bytes_per_second": 4742392.4652728215,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 51.494152404477326,
+    "estimated_full_shards_per_hour": 49.461503889314784,
     "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": 1.1513970990017697,
-      "bytes": 2421895,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 0.9733352420007577,
       "bytes": 6126927,
@@ -562,6 +556,12 @@
       "bytes": 10997948,
       "pages": 27,
       "shards": 3
+    },
+    {
+      "seconds": 1.6622722870015423,
+      "bytes": 7222319,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -403,3 +403,4 @@
 {"step": 760300, "loss": 0.5664460062980652, "lr": 5e-05, "grad_norm": 0.26154402261595466, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111285757983, "tokens_per_second": 152545.6794216428, "data_wait_s": 1.7851001757662743e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 760400, "loss": 0.5658713579177856, "lr": 5e-05, "grad_norm": 0.2475708808148676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111300311346, "tokens_per_second": 112372.13762218344, "data_wait_s": 1.3682001736015081e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 760500, "loss": 0.5541071891784668, "lr": 5e-05, "grad_norm": 0.2431569137163541, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111314904726, "tokens_per_second": 151159.36517041238, "data_wait_s": 1.4037999790161848e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 760600, "loss": 0.5564252734184265, "lr": 5e-05, "grad_norm": 0.24219025165694613, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111329597564, "tokens_per_second": 151954.04795665306, "data_wait_s": 1.232800423167646e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ffcce2765854b46e0629ecd8ba53945e86e7813dd60607f96e419d3d96c14b06
-size 864219
+oid sha256:435cdb0d3a3fa1be1f63f2f8dbc0cab3a9d6988709d71df54f7f4ef6edee3fed
+size 866337