Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75130.72775948797,
-      "required_pages_per_second_upper_estimate": 0.006512141065103789
+      "required_token_positions_per_second": 73738.00369396515,
+      "required_pages_per_second_upper_estimate": 0.006391423272931143
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9391.340969935996,
-      "required_pages_per_second_upper_estimate": 0.003746346325967766
+      "required_token_positions_per_second": 9217.250461745643,
+      "required_pages_per_second_upper_estimate": 0.003676899019365583
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18782.681939871993,
-      "required_pages_per_second_upper_estimate": 0.004345042902877403
+      "required_token_positions_per_second": 18434.500923491287,
+      "required_pages_per_second_upper_estimate": 0.004264497352514326
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28174.022909807987,
-      "required_pages_per_second_upper_estimate": 0.0017156501775330505
+      "required_token_positions_per_second": 27651.75138523693,
+      "required_pages_per_second_upper_estimate": 0.0016838465818335927
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28174.022909807987,
-      "required_pages_per_second_upper_estimate": 0.002168914274284059
+      "required_token_positions_per_second": 27651.75138523693,
+      "required_pages_per_second_upper_estimate": 0.002128708366582468
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28174.022909807987,
-      "required_pages_per_second_upper_estimate": 0.3498270721507877
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27651.75138523693,
+      "required_pages_per_second_upper_estimate": 0.3433422077459668
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.6583513946171777,
-    "producer_p95_seconds": 0.20403043700025592,
-    "gpu_update_seconds": 0.981329506563833,
-    "producer_batches_per_second": 1.518945669708023,
-    "gpu_batches_per_second": 1.0190257128836802,
-    "producer_headroom": 1.4905862045518452,
+    "producer_batch_seconds": 0.09398022477344625,
+    "producer_p95_seconds": 0.19544216100075573,
+    "gpu_update_seconds": 0.999864334624427,
+    "producer_batches_per_second": 10.640536372525746,
+    "gpu_batches_per_second": 1.0001356837831643,
+    "producer_headroom": 10.63909282016247,
     "sustainable": true,
-    "buffer_seconds": 60.84242940695765,
+    "buffer_seconds": 60.99172441209005,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 313491.0801867345,
-    "upload_bytes_per_second": 3460439.2296192227,
+    "generation_bytes_per_second": 313444.0141274131,
+    "upload_bytes_per_second": 3527862.4695014767,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.26788810922172,
+    "estimated_full_shards_per_hour": 67.25778882853312,
     "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": [
@@ -514,6 +514,12 @@
       "bytes": 6465970,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.3629516780001723,
+      "bytes": 11728657,
+      "pages": 30,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -67,3 +67,4 @@
 {"step": 243800, "loss": 0.6316460017114878, "lr": 0.0003, "grad_norm": 0.17855176982115806, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35685005397, "tokens_per_second": 150819.9502725219, "data_wait_s": 1.1596000149438623e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 243900, "loss": 0.5827499339357018, "lr": 0.0003, "grad_norm": 0.14167478314197127, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35699753913, "tokens_per_second": 150785.18489698856, "data_wait_s": 8.988999979919754e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 244000, "loss": 0.5755242705345154, "lr": 0.0003, "grad_norm": 0.15427059476403895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35714566996, "tokens_per_second": 136484.57407254176, "data_wait_s": 1.560100008646259e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 244100, "loss": 0.6154364254325628, "lr": 0.0003, "grad_norm": 0.14847676065465046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35729072522, "tokens_per_second": 147118.2182821808, "data_wait_s": 1.226599943038309e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:91219b9b896b660e0a0682371a68347326154a56c4d7842487d244931f04e899
-size 147090
+oid sha256:45ba531356dd9a68587e1ef552a6a4d08d15ca9f371e931d1fd40b026882e268
+size 149208