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": 77208.71201040965,
-      "required_pages_per_second_upper_estimate": 0.005341477845773905
+      "required_token_positions_per_second": 77380.31348769818,
+      "required_pages_per_second_upper_estimate": 0.005353349634143009
     },
     {
       "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": 9651.089001301207,
-      "required_pages_per_second_upper_estimate": 0.004040709427188857
+      "required_token_positions_per_second": 9672.539185962272,
+      "required_pages_per_second_upper_estimate": 0.004049690171575654
     },
     {
       "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": 19302.178002602413,
-      "required_pages_per_second_upper_estimate": 0.00402013464527719
+      "required_token_positions_per_second": 19345.078371924545,
+      "required_pages_per_second_upper_estimate": 0.00402906966084817
     },
     {
       "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": 28953.267003903624,
-      "required_pages_per_second_upper_estimate": 0.001735704147805096
+      "required_token_positions_per_second": 29017.617557886813,
+      "required_pages_per_second_upper_estimate": 0.0017395618652587817
     },
     {
       "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": 28953.267003903624,
-      "required_pages_per_second_upper_estimate": 0.002250989157590927
+      "required_token_positions_per_second": 29017.617557886813,
+      "required_pages_per_second_upper_estimate": 0.0022559921301149462
     },
     {
       "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": 28953.267003903624,
-      "required_pages_per_second_upper_estimate": 0.3713814215300423
+      "required_token_positions_per_second": 29017.617557886813,
+      "required_pages_per_second_upper_estimate": 0.372206841342297
     }
   ],
   "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": 0.05618330788270498,
-    "producer_p95_seconds": 0.18965457699960098,
-    "gpu_update_seconds": 0.9549181443417891,
-    "producer_batches_per_second": 17.798880800819347,
-    "gpu_batches_per_second": 1.0472101780925789,
-    "producer_headroom": 16.99647422567911,
+    "producer_batch_seconds": 0.07973741292192926,
+    "producer_p95_seconds": 0.210751735001395,
+    "gpu_update_seconds": 0.9528004821500392,
+    "producer_batches_per_second": 12.541164346266136,
+    "gpu_batches_per_second": 1.0495376720879201,
+    "producer_headroom": 11.949227435845255,
     "sustainable": true,
-    "buffer_seconds": 60.159843093532714,
+    "buffer_seconds": 60.02643037545247,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 274534.2711891021,
-    "upload_bytes_per_second": 4735193.854582498,
+    "generation_bytes_per_second": 292377.11323981517,
+    "upload_bytes_per_second": 4693737.876726403,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.908663766429875,
+    "estimated_full_shards_per_hour": 62.73732231040803,
     "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": 1.218460234000304,
-      "bytes": 6273746,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.1453777450001326,
       "bytes": 12048299,
@@ -562,6 +556,12 @@
       "bytes": 11778233,
       "pages": 33,
       "shards": 4
+    },
+    {
+      "seconds": 0.9305929850015673,
+      "bytes": 1085853,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -96,3 +96,4 @@
 {"step": 534600, "loss": 0.5483664274215698, "lr": 0.0003, "grad_norm": 0.2053076472663471, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78248385348, "tokens_per_second": 151455.90838893512, "data_wait_s": 1.0646001101122238e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 534700, "loss": 0.5623101592063904, "lr": 0.0003, "grad_norm": 0.1574369990903295, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78262864506, "tokens_per_second": 150822.15248611808, "data_wait_s": 1.3270000636111945e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 534800, "loss": 0.5913066267967224, "lr": 0.0003, "grad_norm": 0.16676529798441295, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78277489012, "tokens_per_second": 152946.81677670413, "data_wait_s": 1.048099875333719e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 534900, "loss": 0.5911183953285217, "lr": 0.0003, "grad_norm": 0.15948313912046233, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78292146458, "tokens_per_second": 152711.8416193403, "data_wait_s": 1.7186999684781767e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:35e11531172ddaaa4a0ed9e25e5d4da315e3fc20b191633ff15320f3d6ce60f3
-size 208512
+oid sha256:5d988e464b96af34e83e56145c0a106dee91832eee852610dc4436eafa0479ca
+size 210630