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": 76821.33098971241,
-      "required_pages_per_second_upper_estimate": 0.005314677927914265
+      "required_token_positions_per_second": 77103.85088368975,
+      "required_pages_per_second_upper_estimate": 0.005334223309716086
     },
     {
       "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": 9602.666373714052,
-      "required_pages_per_second_upper_estimate": 0.004020435884197564
+      "required_token_positions_per_second": 9637.981360461219,
+      "required_pages_per_second_upper_estimate": 0.004035221531687821
     },
     {
       "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": 19205.332747428103,
-      "required_pages_per_second_upper_estimate": 0.003999964332605508
+      "required_token_positions_per_second": 19275.962720922438,
+      "required_pages_per_second_upper_estimate": 0.0040146746934467195
     },
     {
       "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": 28807.999121142155,
-      "required_pages_per_second_upper_estimate": 0.0017269955600447595
+      "required_token_positions_per_second": 28913.944081383655,
+      "required_pages_per_second_upper_estimate": 0.00173334679864265
     },
     {
       "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": 28807.999121142155,
-      "required_pages_per_second_upper_estimate": 0.002239695218672111
+      "required_token_positions_per_second": 28913.944081383655,
+      "required_pages_per_second_upper_estimate": 0.002247931973328256
     },
     {
       "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": 28807.999121142155,
-      "required_pages_per_second_upper_estimate": 0.3695180811064783
+      "required_token_positions_per_second": 28913.944081383655,
+      "required_pages_per_second_upper_estimate": 0.37087702930162075
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09336983519533959,
-    "producer_p95_seconds": 0.16073739300009038,
-    "gpu_update_seconds": 0.959733436665831,
-    "producer_batches_per_second": 10.710097087650354,
-    "gpu_batches_per_second": 1.0419559867311254,
-    "producer_headroom": 10.27883828495538,
+    "producer_batch_seconds": 0.07292768998436827,
+    "producer_p95_seconds": 0.1573766430001342,
+    "gpu_update_seconds": 0.9562168316497942,
+    "producer_batches_per_second": 13.71221274408041,
+    "gpu_batches_per_second": 1.0457879080361565,
+    "producer_headroom": 13.111848625052502,
     "sustainable": true,
-    "buffer_seconds": 60.463206509947355,
+    "buffer_seconds": 60.24166039393703,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340351.2928608469,
-    "upload_bytes_per_second": 4837391.818290423,
+    "generation_bytes_per_second": 340702.10496351187,
+    "upload_bytes_per_second": 4561176.12483177,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.0314644753366,
+    "estimated_full_shards_per_hour": 73.10674058608072,
     "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": [
@@ -400,6 +400,12 @@
       "bytes": 6556297,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 3.363522853000177,
+      "bytes": 2328587,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -42,3 +42,4 @@
 {"step": 401300, "loss": 0.5392239689826965, "lr": 0.0003, "grad_norm": 0.19113700635905972, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58739160213, "tokens_per_second": 153528.80219505506, "data_wait_s": 1.018900002236478e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 401400, "loss": 0.5845156908035278, "lr": 0.0003, "grad_norm": 0.13922285045245084, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58753818880, "tokens_per_second": 152633.4152228663, "data_wait_s": 9.853999472397845e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 401500, "loss": 0.5698639750480652, "lr": 0.0003, "grad_norm": 0.17107366726336476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58768362398, "tokens_per_second": 151728.81234639054, "data_wait_s": 1.93070000022999e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 401600, "loss": 0.6778021454811096, "lr": 0.0003, "grad_norm": 0.21524274919779954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58783114822, "tokens_per_second": 149674.44077909985, "data_wait_s": 1.2198000149510335e-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 sha256:115a732327b3cf1e2546fb63bddddde9e1b99d39a736ef8e8e6b79f7dae83c6c
-size 94140
+oid sha256:2feb903f7816709d477ac53b10b459c7e0cbc93f7a4be5bd9796a68915facb35
+size 96258