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": 76122.78800998902,
-      "required_pages_per_second_upper_estimate": 0.005266351103733979
+      "required_token_positions_per_second": 75852.86884136873,
+      "required_pages_per_second_upper_estimate": 0.005247677469350069
     },
     {
       "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": 9515.348501248627,
-      "required_pages_per_second_upper_estimate": 0.003983877714400815
+      "required_token_positions_per_second": 9481.608605171092,
+      "required_pages_per_second_upper_estimate": 0.0039697515244823
     },
     {
       "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": 19030.697002497254,
-      "required_pages_per_second_upper_estimate": 0.003963592312390709
+      "required_token_positions_per_second": 18963.217210342184,
+      "required_pages_per_second_upper_estimate": 0.003949538051246598
     },
     {
       "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": 28546.045503745878,
-      "required_pages_per_second_upper_estimate": 0.001711291840661867
+      "required_token_positions_per_second": 28444.825815513275,
+      "required_pages_per_second_upper_estimate": 0.0017052238749058375
     },
     {
       "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": 28546.045503745878,
-      "required_pages_per_second_upper_estimate": 0.0022193294771317438
+      "required_token_positions_per_second": 28444.825815513275,
+      "required_pages_per_second_upper_estimate": 0.0022114600915900152
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28546.045503745878,
-      "required_pages_per_second_upper_estimate": 0.3661580213663996
+      "required_token_positions_per_second": 28444.825815513275,
+      "required_pages_per_second_upper_estimate": 0.3648596838869855
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.06699381209364219,
-    "producer_p95_seconds": 0.2311606559997017,
-    "gpu_update_seconds": 0.9685404584803862,
-    "producer_batches_per_second": 14.926751721520583,
-    "gpu_batches_per_second": 1.0324813911945123,
-    "producer_headroom": 14.457162955984439,
+    "producer_batch_seconds": 0.09371334268766418,
+    "producer_p95_seconds": 0.30137541300064186,
+    "gpu_update_seconds": 0.9719869680102347,
+    "producer_batches_per_second": 10.670839085666652,
+    "gpu_batches_per_second": 1.028820378165266,
+    "producer_headroom": 10.371916529002233,
     "sustainable": true,
-    "buffer_seconds": 60.04950842578394,
+    "buffer_seconds": 60.26319201663455,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 403752.13096224185,
-    "upload_bytes_per_second": 5602505.963495252,
+    "generation_bytes_per_second": 404367.4933112542,
+    "upload_bytes_per_second": 5552437.450134637,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.6358084359211,
+    "estimated_full_shards_per_hour": 86.76785087111682,
     "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": 1.168341065000277,
-      "bytes": 9926928,
-      "pages": 30,
-      "shards": 3
-    },
-    {
-      "seconds": 0.8547535799989419,
-      "bytes": 1160518,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 0.9846142869992036,
       "bytes": 7709639,
@@ -562,6 +550,18 @@
       "bytes": 1121250,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 0.9384285160012951,
+      "bytes": 5824612,
+      "pages": 28,
+      "shards": 2
+    },
+    {
+      "seconds": 0.8717172680007934,
+      "bytes": 283568,
+      "pages": 8,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -177,3 +177,4 @@
 {"step": 414800, "loss": 0.5885172486305237, "lr": 0.0003, "grad_norm": 0.15518945816911042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60715829472, "tokens_per_second": 152416.87492269906, "data_wait_s": 1.551599780214019e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 414900, "loss": 0.5561436414718628, "lr": 0.0003, "grad_norm": 0.14906705278176877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60730466531, "tokens_per_second": 150980.5050675567, "data_wait_s": 6.3920015236362815e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 415000, "loss": 0.6023680567741394, "lr": 0.0003, "grad_norm": 0.15095000167774822, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60745164755, "tokens_per_second": 151931.7864307439, "data_wait_s": 1.166299989563413e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 415100, "loss": 0.5824700593948364, "lr": 0.0003, "grad_norm": 0.15353316396598302, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60759930315, "tokens_per_second": 117552.91918434553, "data_wait_s": 1.4602999726776034e-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 sha256:0a7e6a02d433d791874e8353c5b041ed31ce92c135a92c30474695a02066fdc3
-size 382419
+oid sha256:3219e2666dda796727e31fdb90d8e3885a95c151fcaacdb1be25a8df6aab29cb
+size 384537