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": 77255.12765434662,
-      "required_pages_per_second_upper_estimate": 0.0053446889877205965
+      "required_token_positions_per_second": 77379.42874112996,
+      "required_pages_per_second_upper_estimate": 0.005353288425322525
     },
     {
       "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": 9656.890956793328,
-      "required_pages_per_second_upper_estimate": 0.0040431385847948
+      "required_token_positions_per_second": 9672.428592641245,
+      "required_pages_per_second_upper_estimate": 0.004049643868461591
     },
     {
       "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": 19313.781913586656,
-      "required_pages_per_second_upper_estimate": 0.004022551433919497
+      "required_token_positions_per_second": 19344.85718528249,
+      "required_pages_per_second_upper_estimate": 0.0040290235935037146
     },
     {
       "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": 28970.672870379985,
-      "required_pages_per_second_upper_estimate": 0.0017367476029231347
+      "required_token_positions_per_second": 29017.28577792374,
+      "required_pages_per_second_upper_estimate": 0.0017395419755565939
     },
     {
       "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": 28970.672870379985,
-      "required_pages_per_second_upper_estimate": 0.0022523423871491483
+      "required_token_positions_per_second": 29017.28577792374,
+      "required_pages_per_second_upper_estimate": 0.002255966335682163
     },
     {
       "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": 28970.672870379985,
-      "required_pages_per_second_upper_estimate": 0.3716046852962377
+      "required_token_positions_per_second": 29017.28577792374,
+      "required_pages_per_second_upper_estimate": 0.3722025856251682
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.44007020852345136,
-    "producer_p95_seconds": 0.22791605799829995,
-    "gpu_update_seconds": 0.9543444200865523,
-    "producer_batches_per_second": 2.2723646832519226,
-    "gpu_batches_per_second": 1.0478397305548317,
-    "producer_headroom": 2.168618555863218,
+    "producer_batch_seconds": 0.13764618162502984,
+    "producer_p95_seconds": 0.20426504100032616,
+    "gpu_update_seconds": 0.9528113763498347,
+    "producer_batches_per_second": 7.265003563441807,
+    "gpu_batches_per_second": 1.0495256719445796,
+    "producer_headroom": 6.922178044469442,
     "sustainable": true,
-    "buffer_seconds": 60.1236984654528,
+    "buffer_seconds": 60.027116710039586,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 386777.67616881843,
-    "upload_bytes_per_second": 5626484.658701362,
+    "generation_bytes_per_second": 393558.747603108,
+    "upload_bytes_per_second": 5545390.910560467,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.99348558233656,
+    "estimated_full_shards_per_hour": 84.44854565687113,
     "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": 1.3622697179998795,
-      "bytes": 10825600,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.2178643509996618,
       "bytes": 7619131,
@@ -562,6 +556,12 @@
       "bytes": 10599364,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.9668768979991,
+      "bytes": 2353349,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -153,3 +153,4 @@
 {"step": 572400, "loss": 0.5954951047897339, "lr": 0.0003, "grad_norm": 0.16859362712613973, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83782520139, "tokens_per_second": 151870.59844540505, "data_wait_s": 1.3560000297729857e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 572500, "loss": 0.5691136717796326, "lr": 0.0003, "grad_norm": 0.14838358952968148, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83797088863, "tokens_per_second": 143467.99711045585, "data_wait_s": 7.4670006142696366e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 572600, "loss": 0.6073922514915466, "lr": 0.0003, "grad_norm": 0.17427609695454738, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83811616236, "tokens_per_second": 151883.46306007996, "data_wait_s": 8.850000085658394e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 572700, "loss": 0.6181311011314392, "lr": 0.0003, "grad_norm": 0.18139924881501968, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83826301624, "tokens_per_second": 153689.98038245426, "data_wait_s": 1.3373000911087729e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b05d664374842a793a31e3f115f93869996e322f9d63e866cc48db752e97461a
-size 330804
+oid sha256:352eb41d769b0bb8c4e191ea1eb19be18b157076ee84562bf91e2c24b9aec9de
+size 332922