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": 76082.28891888399,
-      "required_pages_per_second_upper_estimate": 0.0052635492826404945
+      "required_token_positions_per_second": 76487.03253042119,
+      "required_pages_per_second_upper_estimate": 0.005291550384821204
     },
     {
       "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": 9510.286114860499,
-      "required_pages_per_second_upper_estimate": 0.003981758198934754
+      "required_token_positions_per_second": 9560.879066302648,
+      "required_pages_per_second_upper_estimate": 0.004002940411202617
     },
     {
       "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": 19020.572229720998,
-      "required_pages_per_second_upper_estimate": 0.003961483589229629
+      "required_token_positions_per_second": 19121.758132605297,
+      "required_pages_per_second_upper_estimate": 0.003982557944348724
     },
     {
       "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": 28530.8583445815,
-      "required_pages_per_second_upper_estimate": 0.0017103813936594117
+      "required_token_positions_per_second": 28682.637198907945,
+      "required_pages_per_second_upper_estimate": 0.0017194803042233906
     },
     {
       "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": 28530.8583445815,
-      "required_pages_per_second_upper_estimate": 0.0022181487423079665
+      "required_token_positions_per_second": 28682.637198907945,
+      "required_pages_per_second_upper_estimate": 0.0022299488806272224
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28530.8583445815,
-      "required_pages_per_second_upper_estimate": 0.36596321679533994
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28682.637198907945,
+      "required_pages_per_second_upper_estimate": 0.3679100729712029
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.26121670339844627,
-    "producer_p95_seconds": 0.294489550000435,
-    "gpu_update_seconds": 0.9690560187878411,
-    "producer_batches_per_second": 3.8282391094824146,
-    "gpu_batches_per_second": 1.0319320871159396,
-    "producer_headroom": 3.709778150402939,
+    "producer_batch_seconds": 0.07717015535146743,
+    "producer_p95_seconds": 0.27395395299754455,
+    "gpu_update_seconds": 0.9639281007623896,
+    "producer_batches_per_second": 12.958377438085389,
+    "gpu_batches_per_second": 1.0374217736873532,
+    "producer_headroom": 12.490944152855848,
     "sustainable": true,
-    "buffer_seconds": 60.081473164846145,
+    "buffer_seconds": 60.72747034803054,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 437109.7618605049,
-    "upload_bytes_per_second": 5256297.671680878,
+    "generation_bytes_per_second": 439850.77547130577,
+    "upload_bytes_per_second": 5234273.904726351,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.79357950078355,
+    "estimated_full_shards_per_hour": 94.38173721413021,
     "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": 1.5331677430003765,
-      "bytes": 11639669,
-      "pages": 33,
-      "shards": 4
-    },
     {
       "seconds": 0.9482507379998424,
       "bytes": 1158257,
@@ -562,6 +556,12 @@
       "bytes": 2680827,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.5306445749993145,
+      "bytes": 9835087,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -235,3 +235,4 @@
 {"step": 743500, "loss": 0.5424144268035889, "lr": 5e-05, "grad_norm": 0.24468611795184939, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108828331762, "tokens_per_second": 151985.0482702344, "data_wait_s": 1.322199750575237e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 743600, "loss": 0.5389540791511536, "lr": 5e-05, "grad_norm": 0.2382531738229462, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108842916694, "tokens_per_second": 149502.89486380553, "data_wait_s": 1.1493000783957541e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 743700, "loss": 0.5052394866943359, "lr": 5e-05, "grad_norm": 0.2298700973910351, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108857464980, "tokens_per_second": 150381.66727331592, "data_wait_s": 1.0370997188147157e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 743800, "loss": 0.5453187227249146, "lr": 5e-05, "grad_norm": 0.22436211506852036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108871934168, "tokens_per_second": 147656.21708761793, "data_wait_s": 1.6111000149976462e-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 sha256:77a94623105f570f8031fb55f840623542e664396efd6810343672a25924921d
-size 505263
+oid sha256:3f8806bb22f31010f26fed813808bcdee406365b1fcfacf115e62e02af48d19e
+size 507381