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": 76522.96074520277,
-      "required_pages_per_second_upper_estimate": 0.005294035982084738
+      "required_token_positions_per_second": 77410.91467994478,
+      "required_pages_per_second_upper_estimate": 0.005355466695627178
     },
     {
       "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": 9565.370093150346,
-      "required_pages_per_second_upper_estimate": 0.004004820710360443
+      "required_token_positions_per_second": 9676.364334993097,
+      "required_pages_per_second_upper_estimate": 0.004051291681596665
     },
     {
       "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": 19130.740186300693,
-      "required_pages_per_second_upper_estimate": 0.00398442866926079
+      "required_token_positions_per_second": 19352.728669986194,
+      "required_pages_per_second_upper_estimate": 0.0040306630161824845
     },
     {
       "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": 28696.110279451037,
-      "required_pages_per_second_upper_estimate": 0.0017202879948297462
+      "required_token_positions_per_second": 29029.09300497929,
+      "required_pages_per_second_upper_estimate": 0.0017402498007899824
     },
     {
       "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": 28696.110279451037,
-      "required_pages_per_second_upper_estimate": 0.0022309963533776295
+      "required_token_positions_per_second": 29029.09300497929,
+      "required_pages_per_second_upper_estimate": 0.002256884295651229
     },
     {
       "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": 28696.110279451037,
-      "required_pages_per_second_upper_estimate": 0.3680828911821428
+      "required_token_positions_per_second": 29029.09300497929,
+      "required_pages_per_second_upper_estimate": 0.37235403605622414
     }
   ],
   "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": 0.12351496708592435,
-    "producer_p95_seconds": 0.25316149100035545,
-    "gpu_update_seconds": 0.9634755278940513,
-    "producer_batches_per_second": 8.096184807338696,
-    "gpu_batches_per_second": 1.037909081288015,
-    "producer_headroom": 7.8004759311784495,
+    "producer_batch_seconds": 0.09849926410151966,
+    "producer_p95_seconds": 0.24068861799969454,
+    "gpu_update_seconds": 0.9524238320245695,
+    "producer_batches_per_second": 10.152360112755115,
+    "gpu_batches_per_second": 1.0499527273212996,
+    "producer_headroom": 9.669349722683618,
     "sustainable": true,
-    "buffer_seconds": 60.69895825732523,
+    "buffer_seconds": 60.002701417547875,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 351080.9434537606,
-    "upload_bytes_per_second": 4939812.266767675,
+    "generation_bytes_per_second": 385391.72300610965,
+    "upload_bytes_per_second": 4944721.188552387,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.33379771909345,
+    "estimated_full_shards_per_hour": 82.69609229695766,
     "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": 1.0779387669990683,
-      "bytes": 2658914,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 0.9456378320010117,
       "bytes": 5728053,
@@ -562,6 +556,12 @@
       "bytes": 6205762,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.9200315849993785,
+      "bytes": 2298271,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -221,3 +221,4 @@
 {"step": 419200, "loss": 0.5675833821296692, "lr": 0.0003, "grad_norm": 0.1525742805133259, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61359906081, "tokens_per_second": 150730.13709377113, "data_wait_s": 1.671099744271487e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 419300, "loss": 0.6181345582008362, "lr": 0.0003, "grad_norm": 0.1895822870259062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61374469007, "tokens_per_second": 151222.4293424605, "data_wait_s": 1.4278000890044495e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 419400, "loss": 0.5588814616203308, "lr": 0.0003, "grad_norm": 0.18306690284279858, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61388976968, "tokens_per_second": 151427.83731334383, "data_wait_s": 8.581999281886965e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 419500, "loss": 0.5651183724403381, "lr": 0.0003, "grad_norm": 0.15977418495314036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61403576810, "tokens_per_second": 150821.27804004648, "data_wait_s": 1.0966999980155379e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:093d313c6f0eab2e0e1888f296fa01dce751ec69b3533595f8f384f5a73f8a69
-size 475611
+oid sha256:f8c97c8a47b687b0551290e4e4a229a4a7db854fc08cadcbdeac9415896689c5
+size 477729