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": 77752.06238511509,
-      "required_pages_per_second_upper_estimate": 0.005379068085442593
+      "required_token_positions_per_second": 77236.27001614372,
+      "required_pages_per_second_upper_estimate": 0.005343384372553971
     },
     {
       "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": 9719.007798139386,
-      "required_pages_per_second_upper_estimate": 0.004069145609119244
+      "required_token_positions_per_second": 9654.533752017966,
+      "required_pages_per_second_upper_estimate": 0.004042151672379389
     },
     {
       "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": 19438.015596278772,
-      "required_pages_per_second_upper_estimate": 0.004048426033761733
+      "required_token_positions_per_second": 19309.06750403593,
+      "required_pages_per_second_upper_estimate": 0.004021569546737421
     },
     {
       "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": 29157.023394418156,
-      "required_pages_per_second_upper_estimate": 0.0017479190322984483
+      "required_token_positions_per_second": 28963.6012560539,
+      "required_pages_per_second_upper_estimate": 0.0017363236704420156
     },
     {
       "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": 29157.023394418156,
-      "required_pages_per_second_upper_estimate": 0.0022668303207237866
+      "required_token_positions_per_second": 28963.6012560539,
+      "required_pages_per_second_upper_estimate": 0.0022517926002400394
     },
     {
       "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": 29157.023394418156,
-      "required_pages_per_second_upper_estimate": 0.3739949897309957
+      "required_token_positions_per_second": 28963.6012560539,
+      "required_pages_per_second_upper_estimate": 0.37151397822057053
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.10391333152335847,
-    "producer_p95_seconds": 0.29883276400505565,
-    "gpu_update_seconds": 0.9482449434565037,
-    "producer_batches_per_second": 9.62340428643857,
-    "gpu_batches_per_second": 1.0545798392078327,
-    "producer_headroom": 9.125344453453017,
+    "producer_batch_seconds": 0.11758774829661434,
+    "producer_p95_seconds": 0.2902567990022362,
+    "gpu_update_seconds": 0.9545774282547508,
+    "producer_batches_per_second": 8.504287346990491,
+    "gpu_batches_per_second": 1.0475839574672272,
+    "producer_headroom": 8.118000744829601,
     "sustainable": true,
-    "buffer_seconds": 60.687676381216235,
+    "buffer_seconds": 60.1383779800493,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 488792.182430507,
-    "upload_bytes_per_second": 4846836.774540369,
+    "generation_bytes_per_second": 426335.95472615794,
+    "upload_bytes_per_second": 4928154.170979426,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.88342384992987,
+    "estimated_full_shards_per_hour": 91.48177129114679,
     "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": 1.5622936799991294,
-      "bytes": 5942154,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 2.6776525460008997,
       "bytes": 7165592,
@@ -562,6 +556,12 @@
       "bytes": 6891211,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.3423998879952705,
+      "bytes": 11990725,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -345,3 +345,4 @@
 {"step": 919500, "loss": 0.549432098865509, "lr": 3e-05, "grad_norm": 0.3622988405517595, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134587754931, "tokens_per_second": 151684.4907799723, "data_wait_s": 1.6425998182967305e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 919600, "loss": 0.6315422058105469, "lr": 3e-05, "grad_norm": 0.5089820714833988, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134602499802, "tokens_per_second": 154365.2126497351, "data_wait_s": 1.3436998415272683e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 919700, "loss": 0.5307971835136414, "lr": 3e-05, "grad_norm": 0.32777856675305933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134617264984, "tokens_per_second": 154415.3387318978, "data_wait_s": 1.4345998351927847e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 919800, "loss": 0.5458102226257324, "lr": 3e-05, "grad_norm": 0.3303582873094365, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134632073688, "tokens_per_second": 154591.87344219937, "data_wait_s": 9.939998562913388e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d7d4d02e3ab475f9cc29ea4ac14409f313dcec897e7eb3e8b78199715391b368
-size 739809
+oid sha256:5e3c3df11dcd821f5e31f9053d163fa7dce778c181099ed6a325eac75d5f9246
+size 741927