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": 76226.1169229518,
-      "required_pages_per_second_upper_estimate": 0.005273499637688854
+      "required_token_positions_per_second": 76709.22441928578,
+      "required_pages_per_second_upper_estimate": 0.005306922135249068
     },
     {
       "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": 9528.264615368975,
-      "required_pages_per_second_upper_estimate": 0.003989285421663871
+      "required_token_positions_per_second": 9588.653052410722,
+      "required_pages_per_second_upper_estimate": 0.004014568799199285
     },
     {
       "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": 19056.52923073795,
-      "required_pages_per_second_upper_estimate": 0.003968972484291576
+      "required_token_positions_per_second": 19177.306104821444,
+      "required_pages_per_second_upper_estimate": 0.00399412712206281
     },
     {
       "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": 28584.793846106924,
-      "required_pages_per_second_upper_estimate": 0.0017136147446211187
+      "required_token_positions_per_second": 28765.959157232166,
+      "required_pages_per_second_upper_estimate": 0.0017244753283996639
     },
     {
       "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": 28584.793846106924,
-      "required_pages_per_second_upper_estimate": 0.002222341990314369
+      "required_token_positions_per_second": 28765.959157232166,
+      "required_pages_per_second_upper_estimate": 0.0022364267963923684
     },
     {
       "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": 28584.793846106924,
-      "required_pages_per_second_upper_estimate": 0.3666550434974785
+      "required_token_positions_per_second": 28765.959157232166,
+      "required_pages_per_second_upper_estimate": 0.36897883758843736
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.27300006673436883,
-    "producer_p95_seconds": 0.3025267820012232,
-    "gpu_update_seconds": 0.9672275458360701,
-    "producer_batches_per_second": 3.663002767589093,
-    "gpu_batches_per_second": 1.033882879271807,
-    "producer_headroom": 3.5429571772859307,
+    "producer_batch_seconds": 0.09933097020322634,
+    "producer_p95_seconds": 0.3002066709996143,
+    "gpu_update_seconds": 0.9611360375254132,
+    "producer_batches_per_second": 10.067353595299116,
+    "gpu_batches_per_second": 1.040435444054983,
+    "producer_headroom": 9.676096342953015,
     "sustainable": true,
-    "buffer_seconds": 60.93533538767242,
+    "buffer_seconds": 60.55157036410103,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 339322.7889448642,
-    "upload_bytes_per_second": 5346631.938425806,
+    "generation_bytes_per_second": 337993.78712782153,
+    "upload_bytes_per_second": 5368308.3274487825,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.81077147731251,
+    "estimated_full_shards_per_hour": 72.52559862495407,
     "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": 0.9240188730000227,
-      "bytes": 2599575,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.3233367040011217,
       "bytes": 5989613,
@@ -562,6 +556,12 @@
       "bytes": 11958930,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 1.4158421129977796,
+      "bytes": 6969172,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -304,3 +304,4 @@
 {"step": 634800, "loss": 0.519115686416626, "lr": 0.0001, "grad_norm": 0.1724831544526778, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92917987928, "tokens_per_second": 153062.55132019846, "data_wait_s": 1.0308001947123557e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 634900, "loss": 0.5280636548995972, "lr": 0.0001, "grad_norm": 0.16408238043882326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92932609169, "tokens_per_second": 135363.00316597815, "data_wait_s": 9.551000403007492e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 635000, "loss": 0.6401439905166626, "lr": 0.0001, "grad_norm": 0.2037641030489929, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92947329339, "tokens_per_second": 152650.16681481132, "data_wait_s": 1.673600127105601e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 635100, "loss": 0.5057151317596436, "lr": 0.0001, "grad_norm": 0.162308567881318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92962080464, "tokens_per_second": 145506.0382751896, "data_wait_s": 1.3736000255448744e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:42e1c8a67d6e2c75299ae14ee6742d580fce4a3ea3db2810a404e5c126c50337
-size 653754
+oid sha256:e8ee39326709a4a8dcaeeccd2bb9a1f5e6f9d1abfcb2a58bfb1d7b23f154886f
+size 655872