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": 76846.82802828989,
-      "required_pages_per_second_upper_estimate": 0.005316441872204335
+      "required_token_positions_per_second": 76790.29427646406,
+      "required_pages_per_second_upper_estimate": 0.005312530736076647
     },
     {
       "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": 9605.853503536237,
-      "required_pages_per_second_upper_estimate": 0.004021770268899274
+      "required_token_positions_per_second": 9598.786784558008,
+      "required_pages_per_second_upper_estimate": 0.0040188115812329625
     },
     {
       "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": 19211.707007072473,
-      "required_pages_per_second_upper_estimate": 0.004001291922788899
+      "required_token_positions_per_second": 19197.573569116015,
+      "required_pages_per_second_upper_estimate": 0.0039983483003863925
     },
     {
       "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": 28817.560510608713,
-      "required_pages_per_second_upper_estimate": 0.0017275687507438834
+      "required_token_positions_per_second": 28796.360353674023,
+      "required_pages_per_second_upper_estimate": 0.0017262978337064141
     },
     {
       "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": 28817.560510608713,
-      "required_pages_per_second_upper_estimate": 0.0022404385746470283
+      "required_token_positions_per_second": 28796.360353674023,
+      "required_pages_per_second_upper_estimate": 0.0022387903556949913
     },
     {
       "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": 28817.560510608713,
-      "required_pages_per_second_upper_estimate": 0.36964072434433515
+      "required_token_positions_per_second": 28796.360353674023,
+      "required_pages_per_second_upper_estimate": 0.369368791494132
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.0856937427187745,
-    "producer_p95_seconds": 0.2950191459967755,
-    "gpu_update_seconds": 0.959415006340382,
-    "producer_batches_per_second": 11.669463466915555,
-    "gpu_batches_per_second": 1.0423018124496786,
-    "producer_headroom": 11.195858366099644,
+    "producer_batch_seconds": 0.10519672390597634,
+    "producer_p95_seconds": 0.29935628899693256,
+    "gpu_update_seconds": 0.9601213368783424,
+    "producer_batches_per_second": 9.505999453878324,
+    "gpu_batches_per_second": 1.0415350243661032,
+    "producer_headroom": 9.12691290402245,
     "sustainable": true,
-    "buffer_seconds": 60.44314539944407,
+    "buffer_seconds": 60.48764422333557,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382295.08247098717,
-    "upload_bytes_per_second": 5149509.72738284,
+    "generation_bytes_per_second": 382760.89839783544,
+    "upload_bytes_per_second": 5109057.263885623,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.03162532422267,
+    "estimated_full_shards_per_hour": 82.13157857848451,
     "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": 1.271756786001788,
-      "bytes": 10114603,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.2068859079990943,
       "bytes": 6031296,
@@ -562,6 +556,12 @@
       "bytes": 7625671,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.115198325001984,
+      "bytes": 5971301,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -362,3 +362,4 @@
 {"step": 756200, "loss": 0.5434004068374634, "lr": 5e-05, "grad_norm": 0.25498685783512454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110686359307, "tokens_per_second": 151903.76877241887, "data_wait_s": 1.4113997167441994e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 756300, "loss": 0.5599837303161621, "lr": 5e-05, "grad_norm": 0.24912347089497042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110701043673, "tokens_per_second": 152258.57006895452, "data_wait_s": 1.3857999874744564e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 756400, "loss": 0.5434073805809021, "lr": 5e-05, "grad_norm": 0.24177317123749845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110715742977, "tokens_per_second": 152185.61723199894, "data_wait_s": 1.2465003237593919e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 756500, "loss": 0.5631418824195862, "lr": 5e-05, "grad_norm": 0.2474035999962852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110730424511, "tokens_per_second": 150820.3350710897, "data_wait_s": 1.3278993719723076e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8f83977c16388b5a10217420d8060cb8b01cc35acca9aea7143ac95156f880a9
-size 776598
+oid sha256:02536d0b6a2527300a02551d0fb7773988589e0177e8bfe05553592a0999c090
+size 778716