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": 77468.06558313007,
-      "required_pages_per_second_upper_estimate": 0.005359420527717902
+      "required_token_positions_per_second": 77006.21073562939,
+      "required_pages_per_second_upper_estimate": 0.005327468337716914
     },
     {
       "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": 9683.508197891259,
-      "required_pages_per_second_upper_estimate": 0.004054282667811304
+      "required_token_positions_per_second": 9625.776341953673,
+      "required_pages_per_second_upper_estimate": 0.00403011154530848
     },
     {
       "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": 19367.016395782517,
-      "required_pages_per_second_upper_estimate": 0.004033638772673191
+      "required_token_positions_per_second": 19251.552683907346,
+      "required_pages_per_second_upper_estimate": 0.004009590726472442
     },
     {
       "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": 29050.524593673777,
-      "required_pages_per_second_upper_estimate": 0.0017415345918080757
+      "required_token_positions_per_second": 28877.32902586102,
+      "required_pages_per_second_upper_estimate": 0.0017311517819720208
     },
     {
       "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": 29050.524593673777,
-      "required_pages_per_second_upper_estimate": 0.0022585505074046294
+      "required_token_positions_per_second": 28877.32902586102,
+      "required_pages_per_second_upper_estimate": 0.00224508531381398
     },
     {
       "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": 29050.524593673777,
-      "required_pages_per_second_upper_estimate": 0.37262893746454995
+      "required_token_positions_per_second": 28877.32902586102,
+      "required_pages_per_second_upper_estimate": 0.37040737068356
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.12327011946106836,
-    "producer_p95_seconds": 0.29762765200212016,
-    "gpu_update_seconds": 0.9517211956310352,
-    "producer_batches_per_second": 8.112266008761546,
-    "gpu_batches_per_second": 1.0507278860559093,
-    "producer_headroom": 7.7206155051355445,
+    "producer_batch_seconds": 0.1943881238359495,
+    "producer_p95_seconds": 0.30599188699852675,
+    "gpu_update_seconds": 0.9574292683107881,
+    "producer_batches_per_second": 5.144347197074307,
+    "gpu_batches_per_second": 1.0444635787710148,
+    "producer_headroom": 4.925348572831507,
     "sustainable": true,
-    "buffer_seconds": 60.91015652038625,
+    "buffer_seconds": 60.31804390357965,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 174,
+    "upload_pages_per_commit": 160,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 627984.2405698299,
-    "upload_bytes_per_second": 5691075.057239141,
+    "generation_bytes_per_second": 629170.0994676809,
+    "upload_bytes_per_second": 5734629.300267872,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.3753996506747,
+    "estimated_full_shards_per_hour": 67.50262850772295,
     "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": 1.0062806160021864,
-      "bytes": 2394266,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.1657910530011577,
       "bytes": 10819661,
@@ -562,6 +556,12 @@
       "bytes": 7332921,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.8940556989982724,
+      "bytes": 4992530,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -303,3 +303,4 @@
 {"step": 587400, "loss": 0.5683553814888, "lr": 0.0003, "grad_norm": 0.1659644196075371, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85977769325, "tokens_per_second": 152487.73772335792, "data_wait_s": 1.160899773822166e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 587500, "loss": 0.5599024891853333, "lr": 0.0003, "grad_norm": 0.16693459410675637, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85992363447, "tokens_per_second": 151422.57972788348, "data_wait_s": 8.584000170230865e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 587600, "loss": 0.6157953143119812, "lr": 0.0003, "grad_norm": 0.16558787912803208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86007028597, "tokens_per_second": 152591.07657914417, "data_wait_s": 1.0386997018940747e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 587700, "loss": 0.6037683486938477, "lr": 0.0003, "grad_norm": 0.1436797045756794, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86021631913, "tokens_per_second": 152464.6715341517, "data_wait_s": 1.5053999959491193e-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:d269c56d53a2d0c784efe726cd76e7534e9a0850ab73155273371a5c582dfe94
-size 650853
+oid sha256:92a7711f43a3524eed0b4ad1113781049e89a49e977f07d8fb9f1bceb5f18a5f
+size 652971